2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
54 #include "unaligned.h"
56 #include "vlan-bitmap.h"
59 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
61 COVERAGE_DEFINE(ofproto_dpif_expired);
62 COVERAGE_DEFINE(ofproto_dpif_xlate);
63 COVERAGE_DEFINE(facet_changed_rule);
64 COVERAGE_DEFINE(facet_revalidate);
65 COVERAGE_DEFINE(facet_unexpected);
66 COVERAGE_DEFINE(facet_suppress);
68 /* Maximum depth of flow table recursion (due to resubmit actions) in a
69 * flow translation. */
70 #define MAX_RESUBMIT_RECURSION 64
72 /* Number of implemented OpenFlow tables. */
73 enum { N_TABLES = 255 };
74 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
75 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
86 * - Do include packets and bytes from facets that have been deleted or
87 * whose own statistics have been folded into the rule.
89 * - Do include packets and bytes sent "by hand" that were accounted to
90 * the rule without any facet being involved (this is a rare corner
91 * case in rule_execute()).
93 * - Do not include packet or bytes that can be obtained from any facet's
94 * packet_count or byte_count member or that can be obtained from the
95 * datapath by, e.g., dpif_flow_get() for any subfacet.
97 uint64_t packet_count; /* Number of packets received. */
98 uint64_t byte_count; /* Number of bytes received. */
100 tag_type tag; /* Caches rule_calculate_tag() result. */
102 struct list facets; /* List of "struct facet"s. */
105 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
107 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
110 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
111 const struct flow *);
112 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
115 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
116 const struct flow *flow);
118 static void rule_credit_stats(struct rule_dpif *,
119 const struct dpif_flow_stats *);
120 static void flow_push_stats(struct rule_dpif *, const struct flow *,
121 const struct dpif_flow_stats *);
122 static tag_type rule_calculate_tag(const struct flow *,
123 const struct minimask *, uint32_t basis);
124 static void rule_invalidate(const struct rule_dpif *);
126 #define MAX_MIRRORS 32
127 typedef uint32_t mirror_mask_t;
128 #define MIRROR_MASK_C(X) UINT32_C(X)
129 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
131 struct ofproto_dpif *ofproto; /* Owning ofproto. */
132 size_t idx; /* In ofproto's "mirrors" array. */
133 void *aux; /* Key supplied by ofproto's client. */
134 char *name; /* Identifier for log messages. */
136 /* Selection criteria. */
137 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
138 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
139 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
141 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
142 struct ofbundle *out; /* Output port or NULL. */
143 int out_vlan; /* Output VLAN or -1. */
144 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
147 int64_t packet_count; /* Number of packets sent. */
148 int64_t byte_count; /* Number of bytes sent. */
151 static void mirror_destroy(struct ofmirror *);
152 static void update_mirror_stats(struct ofproto_dpif *ofproto,
153 mirror_mask_t mirrors,
154 uint64_t packets, uint64_t bytes);
157 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
158 struct ofproto_dpif *ofproto; /* Owning ofproto. */
159 void *aux; /* Key supplied by ofproto's client. */
160 char *name; /* Identifier for log messages. */
163 struct list ports; /* Contains "struct ofport"s. */
164 enum port_vlan_mode vlan_mode; /* VLAN mode */
165 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
166 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
167 * NULL if all VLANs are trunked. */
168 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
169 struct bond *bond; /* Nonnull iff more than one port. */
170 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
173 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
175 /* Port mirroring info. */
176 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
177 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
178 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
181 static void bundle_remove(struct ofport *);
182 static void bundle_update(struct ofbundle *);
183 static void bundle_destroy(struct ofbundle *);
184 static void bundle_del_port(struct ofport_dpif *);
185 static void bundle_run(struct ofbundle *);
186 static void bundle_wait(struct ofbundle *);
187 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
188 uint16_t in_port, bool warn,
189 struct ofport_dpif **in_ofportp);
191 /* A controller may use OFPP_NONE as the ingress port to indicate that
192 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
193 * when an input bundle is needed for validation (e.g., mirroring or
194 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
195 * any 'port' structs, so care must be taken when dealing with it. */
196 static struct ofbundle ofpp_none_bundle = {
198 .vlan_mode = PORT_VLAN_TRUNK
201 static void stp_run(struct ofproto_dpif *ofproto);
202 static void stp_wait(struct ofproto_dpif *ofproto);
203 static int set_stp_port(struct ofport *,
204 const struct ofproto_port_stp_settings *);
206 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
208 struct action_xlate_ctx {
209 /* action_xlate_ctx_init() initializes these members. */
212 struct ofproto_dpif *ofproto;
214 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
215 * this flow when actions change header fields. */
218 /* The packet corresponding to 'flow', or a null pointer if we are
219 * revalidating without a packet to refer to. */
220 const struct ofpbuf *packet;
222 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
223 * actions update the flow table?
225 * We want to update these tables if we are actually processing a packet,
226 * or if we are accounting for packets that the datapath has processed, but
227 * not if we are just revalidating. */
230 /* The rule that we are currently translating, or NULL. */
231 struct rule_dpif *rule;
233 /* Union of the set of TCP flags seen so far in this flow. (Used only by
234 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
238 /* If nonnull, flow translation calls this function just before executing a
239 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
240 * when the recursion depth is exceeded.
242 * 'rule' is the rule being submitted into. It will be null if the
243 * resubmit or OFPP_TABLE action didn't find a matching rule.
245 * This is normally null so the client has to set it manually after
246 * calling action_xlate_ctx_init(). */
247 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
249 /* If nonnull, flow translation calls this function to report some
250 * significant decision, e.g. to explain why OFPP_NORMAL translation
251 * dropped a packet. */
252 void (*report_hook)(struct action_xlate_ctx *, const char *s);
254 /* If nonnull, flow translation credits the specified statistics to each
255 * rule reached through a resubmit or OFPP_TABLE action.
257 * This is normally null so the client has to set it manually after
258 * calling action_xlate_ctx_init(). */
259 const struct dpif_flow_stats *resubmit_stats;
261 /* xlate_actions() initializes and uses these members. The client might want
262 * to look at them after it returns. */
264 struct ofpbuf *odp_actions; /* Datapath actions. */
265 tag_type tags; /* Tags associated with actions. */
266 enum slow_path_reason slow; /* 0 if fast path may be used. */
267 bool has_learn; /* Actions include NXAST_LEARN? */
268 bool has_normal; /* Actions output to OFPP_NORMAL? */
269 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
270 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
271 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
273 /* xlate_actions() initializes and uses these members, but the client has no
274 * reason to look at them. */
276 int recurse; /* Recursion level, via xlate_table_action. */
277 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
278 struct flow base_flow; /* Flow at the last commit. */
279 uint32_t orig_skb_priority; /* Priority when packet arrived. */
280 uint8_t table_id; /* OpenFlow table ID where flow was found. */
281 uint32_t sflow_n_outputs; /* Number of output ports. */
282 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
283 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
284 bool exit; /* No further actions should be processed. */
285 struct flow orig_flow; /* Copy of original flow. */
288 static void action_xlate_ctx_init(struct action_xlate_ctx *,
289 struct ofproto_dpif *, const struct flow *,
290 ovs_be16 initial_tci, struct rule_dpif *,
291 uint8_t tcp_flags, const struct ofpbuf *);
292 static void xlate_actions(struct action_xlate_ctx *,
293 const struct ofpact *ofpacts, size_t ofpacts_len,
294 struct ofpbuf *odp_actions);
295 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
296 const struct ofpact *ofpacts,
299 static size_t put_userspace_action(const struct ofproto_dpif *,
300 struct ofpbuf *odp_actions,
302 const union user_action_cookie *);
304 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
305 enum slow_path_reason,
306 uint64_t *stub, size_t stub_size,
307 const struct nlattr **actionsp,
308 size_t *actions_lenp);
310 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
312 /* A subfacet (see "struct subfacet" below) has three possible installation
315 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
316 * case just after the subfacet is created, just before the subfacet is
317 * destroyed, or if the datapath returns an error when we try to install a
320 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
322 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
323 * ofproto_dpif is installed in the datapath.
326 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
327 SF_FAST_PATH, /* Full actions are installed. */
328 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
331 static const char *subfacet_path_to_string(enum subfacet_path);
333 /* A dpif flow and actions associated with a facet.
335 * See also the large comment on struct facet. */
338 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
339 struct list list_node; /* In struct facet's 'facets' list. */
340 struct facet *facet; /* Owning facet. */
344 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
345 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
346 * regenerate the ODP flow key from ->facet->flow. */
347 enum odp_key_fitness key_fitness;
351 long long int used; /* Time last used; time created if not used. */
353 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
354 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
358 * These should be essentially identical for every subfacet in a facet, but
359 * may differ in trivial ways due to VLAN splinters. */
360 size_t actions_len; /* Number of bytes in actions[]. */
361 struct nlattr *actions; /* Datapath actions. */
363 enum slow_path_reason slow; /* 0 if fast path may be used. */
364 enum subfacet_path path; /* Installed in datapath? */
366 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
367 * splinters can cause it to differ. This value should be removed when
368 * the VLAN splinters feature is no longer needed. */
369 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
371 /* Datapath port the packet arrived on. This is needed to remove
372 * flows for ports that are no longer part of the bridge. Since the
373 * flow definition only has the OpenFlow port number and the port is
374 * no longer part of the bridge, we can't determine the datapath port
375 * number needed to delete the flow from the datapath. */
376 uint32_t odp_in_port;
379 #define SUBFACET_DESTROY_MAX_BATCH 50
381 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
383 static struct subfacet *subfacet_find(struct ofproto_dpif *,
384 const struct nlattr *key, size_t key_len,
386 const struct flow *flow);
387 static void subfacet_destroy(struct subfacet *);
388 static void subfacet_destroy__(struct subfacet *);
389 static void subfacet_destroy_batch(struct ofproto_dpif *,
390 struct subfacet **, int n);
391 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
393 static void subfacet_reset_dp_stats(struct subfacet *,
394 struct dpif_flow_stats *);
395 static void subfacet_update_time(struct subfacet *, long long int used);
396 static void subfacet_update_stats(struct subfacet *,
397 const struct dpif_flow_stats *);
398 static void subfacet_make_actions(struct subfacet *,
399 const struct ofpbuf *packet,
400 struct ofpbuf *odp_actions);
401 static int subfacet_install(struct subfacet *,
402 const struct nlattr *actions, size_t actions_len,
403 struct dpif_flow_stats *, enum slow_path_reason);
404 static void subfacet_uninstall(struct subfacet *);
406 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
408 /* An exact-match instantiation of an OpenFlow flow.
410 * A facet associates a "struct flow", which represents the Open vSwitch
411 * userspace idea of an exact-match flow, with one or more subfacets. Each
412 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
413 * the facet. When the kernel module (or other dpif implementation) and Open
414 * vSwitch userspace agree on the definition of a flow key, there is exactly
415 * one subfacet per facet. If the dpif implementation supports more-specific
416 * flow matching than userspace, however, a facet can have more than one
417 * subfacet, each of which corresponds to some distinction in flow that
418 * userspace simply doesn't understand.
420 * Flow expiration works in terms of subfacets, so a facet must have at least
421 * one subfacet or it will never expire, leaking memory. */
424 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
425 struct list list_node; /* In owning rule's 'facets' list. */
426 struct rule_dpif *rule; /* Owning rule. */
429 struct list subfacets;
430 long long int used; /* Time last used; time created if not used. */
437 * - Do include packets and bytes sent "by hand", e.g. with
440 * - Do include packets and bytes that were obtained from the datapath
441 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
442 * DPIF_FP_ZERO_STATS).
444 * - Do not include packets or bytes that can be obtained from the
445 * datapath for any existing subfacet.
447 uint64_t packet_count; /* Number of packets received. */
448 uint64_t byte_count; /* Number of bytes received. */
450 /* Resubmit statistics. */
451 uint64_t prev_packet_count; /* Number of packets from last stats push. */
452 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
453 long long int prev_used; /* Used time from last stats push. */
456 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
457 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
458 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
460 /* Properties of datapath actions.
462 * Every subfacet has its own actions because actions can differ slightly
463 * between splintered and non-splintered subfacets due to the VLAN tag
464 * being initially different (present vs. absent). All of them have these
465 * properties in common so we just store one copy of them here. */
466 bool has_learn; /* Actions include NXAST_LEARN? */
467 bool has_normal; /* Actions output to OFPP_NORMAL? */
468 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
469 tag_type tags; /* Tags that would require revalidation. */
470 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
472 /* Storage for a single subfacet, to reduce malloc() time and space
473 * overhead. (A facet always has at least one subfacet and in the common
474 * case has exactly one subfacet.) */
475 struct subfacet one_subfacet;
478 static struct facet *facet_create(struct rule_dpif *,
479 const struct flow *, uint32_t hash);
480 static void facet_remove(struct facet *);
481 static void facet_free(struct facet *);
483 static struct facet *facet_find(struct ofproto_dpif *,
484 const struct flow *, uint32_t hash);
485 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
486 const struct flow *, uint32_t hash);
487 static void facet_revalidate(struct facet *);
488 static bool facet_check_consistency(struct facet *);
490 static void facet_flush_stats(struct facet *);
492 static void facet_update_time(struct facet *, long long int used);
493 static void facet_reset_counters(struct facet *);
494 static void facet_push_stats(struct facet *);
495 static void facet_learn(struct facet *);
496 static void facet_account(struct facet *);
498 static bool facet_is_controller_flow(struct facet *);
501 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
505 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
506 struct list bundle_node; /* In struct ofbundle's "ports" list. */
507 struct cfm *cfm; /* Connectivity Fault Management, if any. */
508 tag_type tag; /* Tag associated with this port. */
509 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
510 bool may_enable; /* May be enabled in bonds. */
511 long long int carrier_seq; /* Carrier status changes. */
514 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
515 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
516 long long int stp_state_entered;
518 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
520 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
522 * This is deprecated. It is only for compatibility with broken device
523 * drivers in old versions of Linux that do not properly support VLANs when
524 * VLAN devices are not used. When broken device drivers are no longer in
525 * widespread use, we will delete these interfaces. */
526 uint16_t realdev_ofp_port;
530 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
531 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
532 * traffic egressing the 'ofport' with that priority should be marked with. */
533 struct priority_to_dscp {
534 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
535 uint32_t priority; /* Priority of this queue (see struct flow). */
537 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
540 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
542 * This is deprecated. It is only for compatibility with broken device drivers
543 * in old versions of Linux that do not properly support VLANs when VLAN
544 * devices are not used. When broken device drivers are no longer in
545 * widespread use, we will delete these interfaces. */
546 struct vlan_splinter {
547 struct hmap_node realdev_vid_node;
548 struct hmap_node vlandev_node;
549 uint16_t realdev_ofp_port;
550 uint16_t vlandev_ofp_port;
554 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
555 uint32_t realdev, ovs_be16 vlan_tci);
556 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
557 static void vsp_remove(struct ofport_dpif *);
558 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
560 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
562 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
565 static struct ofport_dpif *
566 ofport_dpif_cast(const struct ofport *ofport)
568 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
569 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
572 static void port_run(struct ofport_dpif *);
573 static void port_run_fast(struct ofport_dpif *);
574 static void port_wait(struct ofport_dpif *);
575 static int set_cfm(struct ofport *, const struct cfm_settings *);
576 static void ofport_clear_priorities(struct ofport_dpif *);
578 struct dpif_completion {
579 struct list list_node;
580 struct ofoperation *op;
583 /* Extra information about a classifier table.
584 * Currently used just for optimized flow revalidation. */
586 /* If either of these is nonnull, then this table has a form that allows
587 * flows to be tagged to avoid revalidating most flows for the most common
588 * kinds of flow table changes. */
589 struct cls_table *catchall_table; /* Table that wildcards all fields. */
590 struct cls_table *other_table; /* Table with any other wildcard set. */
591 uint32_t basis; /* Keeps each table's tags separate. */
594 /* Reasons that we might need to revalidate every facet, and corresponding
597 * A value of 0 means that there is no need to revalidate.
599 * It would be nice to have some cleaner way to integrate with coverage
600 * counters, but with only a few reasons I guess this is good enough for
602 enum revalidate_reason {
603 REV_RECONFIGURE = 1, /* Switch configuration changed. */
604 REV_STP, /* Spanning tree protocol port status change. */
605 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
606 REV_FLOW_TABLE, /* Flow table changed. */
607 REV_INCONSISTENCY /* Facet self-check failed. */
609 COVERAGE_DEFINE(rev_reconfigure);
610 COVERAGE_DEFINE(rev_stp);
611 COVERAGE_DEFINE(rev_port_toggled);
612 COVERAGE_DEFINE(rev_flow_table);
613 COVERAGE_DEFINE(rev_inconsistency);
615 /* All datapaths of a given type share a single dpif backer instance. */
620 struct timer next_expiration;
621 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
624 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
625 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
627 static struct ofport_dpif *
628 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
630 struct ofproto_dpif {
631 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
633 struct dpif_backer *backer;
635 /* Special OpenFlow rules. */
636 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
637 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
643 struct netflow *netflow;
644 struct dpif_sflow *sflow;
645 struct hmap bundles; /* Contains "struct ofbundle"s. */
646 struct mac_learning *ml;
647 struct ofmirror *mirrors[MAX_MIRRORS];
649 bool has_bonded_bundles;
653 struct hmap subfacets;
654 struct governor *governor;
657 struct table_dpif tables[N_TABLES];
658 enum revalidate_reason need_revalidate;
659 struct tag_set revalidate_set;
661 /* Support for debugging async flow mods. */
662 struct list completions;
664 bool has_bundle_action; /* True when the first bundle action appears. */
665 struct netdev_stats stats; /* To account packets generated and consumed in
670 long long int stp_last_tick;
672 /* VLAN splinters. */
673 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
674 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
677 struct sset ports; /* Set of port names. */
678 struct sset port_poll_set; /* Queued names for port_poll() reply. */
679 int port_poll_errno; /* Last errno for port_poll() reply. */
682 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
683 * for debugging the asynchronous flow_mod implementation.) */
686 /* All existing ofproto_dpif instances, indexed by ->up.name. */
687 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
689 static void ofproto_dpif_unixctl_init(void);
691 static struct ofproto_dpif *
692 ofproto_dpif_cast(const struct ofproto *ofproto)
694 assert(ofproto->ofproto_class == &ofproto_dpif_class);
695 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
698 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
700 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
702 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
703 const struct ofpbuf *, ovs_be16 initial_tci,
706 /* Packet processing. */
707 static void update_learning_table(struct ofproto_dpif *,
708 const struct flow *, int vlan,
711 #define FLOW_MISS_MAX_BATCH 50
712 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
714 /* Flow expiration. */
715 static int expire(struct dpif_backer *);
718 static void send_netflow_active_timeouts(struct ofproto_dpif *);
721 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
722 static size_t compose_sflow_action(const struct ofproto_dpif *,
723 struct ofpbuf *odp_actions,
724 const struct flow *, uint32_t odp_port);
725 static void add_mirror_actions(struct action_xlate_ctx *ctx,
726 const struct flow *flow);
727 /* Global variables. */
728 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
730 /* Initial mappings of port to bridge mappings. */
731 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
733 /* Factory functions. */
736 init(const struct shash *iface_hints)
738 struct shash_node *node;
740 /* Make a local copy, since we don't own 'iface_hints' elements. */
741 SHASH_FOR_EACH(node, iface_hints) {
742 const struct iface_hint *orig_hint = node->data;
743 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
745 new_hint->br_name = xstrdup(orig_hint->br_name);
746 new_hint->br_type = xstrdup(orig_hint->br_type);
747 new_hint->ofp_port = orig_hint->ofp_port;
749 shash_add(&init_ofp_ports, node->name, new_hint);
754 enumerate_types(struct sset *types)
756 dp_enumerate_types(types);
760 enumerate_names(const char *type, struct sset *names)
762 struct ofproto_dpif *ofproto;
765 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
766 if (strcmp(type, ofproto->up.type)) {
769 sset_add(names, ofproto->up.name);
776 del(const char *type, const char *name)
781 error = dpif_open(name, type, &dpif);
783 error = dpif_delete(dpif);
790 port_open_type(const char *datapath_type, const char *port_type)
792 return dpif_port_open_type(datapath_type, port_type);
795 /* Type functions. */
797 static struct ofproto_dpif *
798 lookup_ofproto_dpif_by_port_name(const char *name)
800 struct ofproto_dpif *ofproto;
802 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
803 if (sset_contains(&ofproto->ports, name)) {
812 type_run(const char *type)
814 struct dpif_backer *backer;
818 backer = shash_find_data(&all_dpif_backers, type);
820 /* This is not necessarily a problem, since backers are only
821 * created on demand. */
825 dpif_run(backer->dpif);
827 if (timer_expired(&backer->next_expiration)) {
828 int delay = expire(backer);
829 timer_set_duration(&backer->next_expiration, delay);
832 /* Check for port changes in the dpif. */
833 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
834 struct ofproto_dpif *ofproto;
835 struct dpif_port port;
837 /* Don't report on the datapath's device. */
838 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
842 ofproto = lookup_ofproto_dpif_by_port_name(devname);
843 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
844 /* The port was removed. If we know the datapath,
845 * report it through poll_set(). If we don't, it may be
846 * notifying us of a removal we initiated, so ignore it.
847 * If there's a pending ENOBUFS, let it stand, since
848 * everything will be reevaluated. */
849 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
850 sset_add(&ofproto->port_poll_set, devname);
851 ofproto->port_poll_errno = 0;
853 } else if (!ofproto) {
854 /* The port was added, but we don't know with which
855 * ofproto we should associate it. Delete it. */
856 dpif_port_del(backer->dpif, port.port_no);
858 dpif_port_destroy(&port);
863 if (error != EAGAIN) {
864 struct ofproto_dpif *ofproto;
866 /* There was some sort of error, so propagate it to all
867 * ofprotos that use this backer. */
868 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
869 &all_ofproto_dpifs) {
870 if (ofproto->backer == backer) {
871 sset_clear(&ofproto->port_poll_set);
872 ofproto->port_poll_errno = error;
881 type_run_fast(const char *type)
883 struct dpif_backer *backer;
886 backer = shash_find_data(&all_dpif_backers, type);
888 /* This is not necessarily a problem, since backers are only
889 * created on demand. */
893 /* Handle one or more batches of upcalls, until there's nothing left to do
894 * or until we do a fixed total amount of work.
896 * We do work in batches because it can be much cheaper to set up a number
897 * of flows and fire off their patches all at once. We do multiple batches
898 * because in some cases handling a packet can cause another packet to be
899 * queued almost immediately as part of the return flow. Both
900 * optimizations can make major improvements on some benchmarks and
901 * presumably for real traffic as well. */
903 while (work < FLOW_MISS_MAX_BATCH) {
904 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
915 type_wait(const char *type)
917 struct dpif_backer *backer;
919 backer = shash_find_data(&all_dpif_backers, type);
921 /* This is not necessarily a problem, since backers are only
922 * created on demand. */
926 timer_wait(&backer->next_expiration);
929 /* Basic life-cycle. */
931 static int add_internal_flows(struct ofproto_dpif *);
933 static struct ofproto *
936 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
941 dealloc(struct ofproto *ofproto_)
943 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
948 close_dpif_backer(struct dpif_backer *backer)
950 struct shash_node *node;
952 assert(backer->refcount > 0);
954 if (--backer->refcount) {
958 hmap_destroy(&backer->odp_to_ofport_map);
959 node = shash_find(&all_dpif_backers, backer->type);
961 shash_delete(&all_dpif_backers, node);
962 dpif_close(backer->dpif);
967 /* Datapath port slated for removal from datapath. */
969 struct list list_node;
974 open_dpif_backer(const char *type, struct dpif_backer **backerp)
976 struct dpif_backer *backer;
977 struct dpif_port_dump port_dump;
978 struct dpif_port port;
979 struct shash_node *node;
980 struct list garbage_list;
981 struct odp_garbage *garbage, *next;
987 backer = shash_find_data(&all_dpif_backers, type);
994 backer_name = xasprintf("ovs-%s", type);
996 /* Remove any existing datapaths, since we assume we're the only
997 * userspace controlling the datapath. */
999 dp_enumerate_names(type, &names);
1000 SSET_FOR_EACH(name, &names) {
1001 struct dpif *old_dpif;
1003 /* Don't remove our backer if it exists. */
1004 if (!strcmp(name, backer_name)) {
1008 if (dpif_open(name, type, &old_dpif)) {
1009 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1011 dpif_delete(old_dpif);
1012 dpif_close(old_dpif);
1015 sset_destroy(&names);
1017 backer = xmalloc(sizeof *backer);
1019 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1022 VLOG_ERR("failed to open datapath of type %s: %s", type,
1027 backer->type = xstrdup(type);
1028 backer->refcount = 1;
1029 hmap_init(&backer->odp_to_ofport_map);
1030 timer_set_duration(&backer->next_expiration, 1000);
1033 dpif_flow_flush(backer->dpif);
1035 /* Loop through the ports already on the datapath and remove any
1036 * that we don't need anymore. */
1037 list_init(&garbage_list);
1038 dpif_port_dump_start(&port_dump, backer->dpif);
1039 while (dpif_port_dump_next(&port_dump, &port)) {
1040 node = shash_find(&init_ofp_ports, port.name);
1041 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1042 garbage = xmalloc(sizeof *garbage);
1043 garbage->odp_port = port.port_no;
1044 list_push_front(&garbage_list, &garbage->list_node);
1047 dpif_port_dump_done(&port_dump);
1049 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1050 dpif_port_del(backer->dpif, garbage->odp_port);
1051 list_remove(&garbage->list_node);
1055 shash_add(&all_dpif_backers, type, backer);
1057 error = dpif_recv_set(backer->dpif, true);
1059 VLOG_ERR("failed to listen on datapath of type %s: %s",
1060 type, strerror(error));
1061 close_dpif_backer(backer);
1069 construct(struct ofproto *ofproto_)
1071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1072 struct shash_node *node, *next;
1077 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1082 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1083 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1085 ofproto->n_matches = 0;
1087 ofproto->netflow = NULL;
1088 ofproto->sflow = NULL;
1089 ofproto->stp = NULL;
1090 hmap_init(&ofproto->bundles);
1091 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1092 for (i = 0; i < MAX_MIRRORS; i++) {
1093 ofproto->mirrors[i] = NULL;
1095 ofproto->has_bonded_bundles = false;
1097 hmap_init(&ofproto->facets);
1098 hmap_init(&ofproto->subfacets);
1099 ofproto->governor = NULL;
1101 for (i = 0; i < N_TABLES; i++) {
1102 struct table_dpif *table = &ofproto->tables[i];
1104 table->catchall_table = NULL;
1105 table->other_table = NULL;
1106 table->basis = random_uint32();
1108 ofproto->need_revalidate = 0;
1109 tag_set_init(&ofproto->revalidate_set);
1111 list_init(&ofproto->completions);
1113 ofproto_dpif_unixctl_init();
1115 ofproto->has_mirrors = false;
1116 ofproto->has_bundle_action = false;
1118 hmap_init(&ofproto->vlandev_map);
1119 hmap_init(&ofproto->realdev_vid_map);
1121 sset_init(&ofproto->ports);
1122 sset_init(&ofproto->port_poll_set);
1123 ofproto->port_poll_errno = 0;
1125 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1126 const struct iface_hint *iface_hint = node->data;
1128 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1129 /* Check if the datapath already has this port. */
1130 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1131 sset_add(&ofproto->ports, node->name);
1134 free(iface_hint->br_name);
1135 free(iface_hint->br_type);
1136 shash_delete(&init_ofp_ports, node);
1140 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1141 hash_string(ofproto->up.name, 0));
1142 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1144 ofproto_init_tables(ofproto_, N_TABLES);
1145 error = add_internal_flows(ofproto);
1146 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1152 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1153 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1155 struct ofputil_flow_mod fm;
1158 match_init_catchall(&fm.match);
1160 match_set_reg(&fm.match, 0, id);
1161 fm.new_cookie = htonll(0);
1162 fm.cookie = htonll(0);
1163 fm.cookie_mask = htonll(0);
1164 fm.table_id = TBL_INTERNAL;
1165 fm.command = OFPFC_ADD;
1166 fm.idle_timeout = 0;
1167 fm.hard_timeout = 0;
1171 fm.ofpacts = ofpacts->data;
1172 fm.ofpacts_len = ofpacts->size;
1174 error = ofproto_flow_mod(&ofproto->up, &fm);
1176 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1177 id, ofperr_to_string(error));
1181 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1182 assert(*rulep != NULL);
1188 add_internal_flows(struct ofproto_dpif *ofproto)
1190 struct ofpact_controller *controller;
1191 uint64_t ofpacts_stub[128 / 8];
1192 struct ofpbuf ofpacts;
1196 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1199 controller = ofpact_put_CONTROLLER(&ofpacts);
1200 controller->max_len = UINT16_MAX;
1201 controller->controller_id = 0;
1202 controller->reason = OFPR_NO_MATCH;
1203 ofpact_pad(&ofpacts);
1205 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1210 ofpbuf_clear(&ofpacts);
1211 error = add_internal_flow(ofproto, id++, &ofpacts,
1212 &ofproto->no_packet_in_rule);
1217 complete_operations(struct ofproto_dpif *ofproto)
1219 struct dpif_completion *c, *next;
1221 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1222 ofoperation_complete(c->op, 0);
1223 list_remove(&c->list_node);
1229 destruct(struct ofproto *ofproto_)
1231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1232 struct rule_dpif *rule, *next_rule;
1233 struct oftable *table;
1236 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1237 complete_operations(ofproto);
1239 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1240 struct cls_cursor cursor;
1242 cls_cursor_init(&cursor, &table->cls, NULL);
1243 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1244 ofproto_rule_destroy(&rule->up);
1248 for (i = 0; i < MAX_MIRRORS; i++) {
1249 mirror_destroy(ofproto->mirrors[i]);
1252 netflow_destroy(ofproto->netflow);
1253 dpif_sflow_destroy(ofproto->sflow);
1254 hmap_destroy(&ofproto->bundles);
1255 mac_learning_destroy(ofproto->ml);
1257 hmap_destroy(&ofproto->facets);
1258 hmap_destroy(&ofproto->subfacets);
1259 governor_destroy(ofproto->governor);
1261 hmap_destroy(&ofproto->vlandev_map);
1262 hmap_destroy(&ofproto->realdev_vid_map);
1264 sset_destroy(&ofproto->ports);
1265 sset_destroy(&ofproto->port_poll_set);
1267 close_dpif_backer(ofproto->backer);
1271 run_fast(struct ofproto *ofproto_)
1273 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1274 struct ofport_dpif *ofport;
1276 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1277 port_run_fast(ofport);
1284 run(struct ofproto *ofproto_)
1286 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1287 struct ofport_dpif *ofport;
1288 struct ofbundle *bundle;
1292 complete_operations(ofproto);
1295 error = run_fast(ofproto_);
1300 if (ofproto->netflow) {
1301 if (netflow_run(ofproto->netflow)) {
1302 send_netflow_active_timeouts(ofproto);
1305 if (ofproto->sflow) {
1306 dpif_sflow_run(ofproto->sflow);
1309 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1312 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1317 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1319 /* Now revalidate if there's anything to do. */
1320 if (ofproto->need_revalidate
1321 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1322 struct tag_set revalidate_set = ofproto->revalidate_set;
1323 bool revalidate_all = ofproto->need_revalidate;
1324 struct facet *facet;
1326 switch (ofproto->need_revalidate) {
1327 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1328 case REV_STP: COVERAGE_INC(rev_stp); break;
1329 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1330 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1331 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1334 /* Clear the revalidation flags. */
1335 tag_set_init(&ofproto->revalidate_set);
1336 ofproto->need_revalidate = 0;
1338 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1340 || tag_set_intersects(&revalidate_set, facet->tags)) {
1341 facet_revalidate(facet);
1346 /* Check the consistency of a random facet, to aid debugging. */
1347 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1348 struct facet *facet;
1350 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1351 struct facet, hmap_node);
1352 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1353 if (!facet_check_consistency(facet)) {
1354 ofproto->need_revalidate = REV_INCONSISTENCY;
1359 if (ofproto->governor) {
1362 governor_run(ofproto->governor);
1364 /* If the governor has shrunk to its minimum size and the number of
1365 * subfacets has dwindled, then drop the governor entirely.
1367 * For hysteresis, the number of subfacets to drop the governor is
1368 * smaller than the number needed to trigger its creation. */
1369 n_subfacets = hmap_count(&ofproto->subfacets);
1370 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1371 && governor_is_idle(ofproto->governor)) {
1372 governor_destroy(ofproto->governor);
1373 ofproto->governor = NULL;
1381 wait(struct ofproto *ofproto_)
1383 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1384 struct ofport_dpif *ofport;
1385 struct ofbundle *bundle;
1387 if (!clogged && !list_is_empty(&ofproto->completions)) {
1388 poll_immediate_wake();
1391 dpif_wait(ofproto->backer->dpif);
1392 dpif_recv_wait(ofproto->backer->dpif);
1393 if (ofproto->sflow) {
1394 dpif_sflow_wait(ofproto->sflow);
1396 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1397 poll_immediate_wake();
1399 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1402 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1403 bundle_wait(bundle);
1405 if (ofproto->netflow) {
1406 netflow_wait(ofproto->netflow);
1408 mac_learning_wait(ofproto->ml);
1410 if (ofproto->need_revalidate) {
1411 /* Shouldn't happen, but if it does just go around again. */
1412 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1413 poll_immediate_wake();
1415 if (ofproto->governor) {
1416 governor_wait(ofproto->governor);
1421 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1423 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1425 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1426 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1430 flush(struct ofproto *ofproto_)
1432 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1433 struct subfacet *subfacet, *next_subfacet;
1434 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1438 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1439 &ofproto->subfacets) {
1440 if (subfacet->path != SF_NOT_INSTALLED) {
1441 batch[n_batch++] = subfacet;
1442 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1443 subfacet_destroy_batch(ofproto, batch, n_batch);
1447 subfacet_destroy(subfacet);
1452 subfacet_destroy_batch(ofproto, batch, n_batch);
1457 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1458 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1460 *arp_match_ip = true;
1461 *actions = (OFPUTIL_A_OUTPUT |
1462 OFPUTIL_A_SET_VLAN_VID |
1463 OFPUTIL_A_SET_VLAN_PCP |
1464 OFPUTIL_A_STRIP_VLAN |
1465 OFPUTIL_A_SET_DL_SRC |
1466 OFPUTIL_A_SET_DL_DST |
1467 OFPUTIL_A_SET_NW_SRC |
1468 OFPUTIL_A_SET_NW_DST |
1469 OFPUTIL_A_SET_NW_TOS |
1470 OFPUTIL_A_SET_TP_SRC |
1471 OFPUTIL_A_SET_TP_DST |
1476 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1478 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1479 struct dpif_dp_stats s;
1481 strcpy(ots->name, "classifier");
1483 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1485 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1486 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1489 static struct ofport *
1492 struct ofport_dpif *port = xmalloc(sizeof *port);
1497 port_dealloc(struct ofport *port_)
1499 struct ofport_dpif *port = ofport_dpif_cast(port_);
1504 port_construct(struct ofport *port_)
1506 struct ofport_dpif *port = ofport_dpif_cast(port_);
1507 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1508 struct dpif_port dpif_port;
1511 ofproto->need_revalidate = REV_RECONFIGURE;
1512 port->bundle = NULL;
1514 port->tag = tag_create_random();
1515 port->may_enable = true;
1516 port->stp_port = NULL;
1517 port->stp_state = STP_DISABLED;
1518 hmap_init(&port->priorities);
1519 port->realdev_ofp_port = 0;
1520 port->vlandev_vid = 0;
1521 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1523 error = dpif_port_query_by_name(ofproto->backer->dpif,
1524 netdev_get_name(port->up.netdev),
1530 port->odp_port = dpif_port.port_no;
1532 /* Sanity-check that a mapping doesn't already exist. This
1533 * shouldn't happen. */
1534 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1535 VLOG_ERR("port %s already has an OpenFlow port number\n",
1540 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1541 hash_int(port->odp_port, 0));
1543 if (ofproto->sflow) {
1544 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1551 port_destruct(struct ofport *port_)
1553 struct ofport_dpif *port = ofport_dpif_cast(port_);
1554 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1555 const char *devname = netdev_get_name(port->up.netdev);
1557 if (dpif_port_exists(ofproto->backer->dpif, devname)) {
1558 /* The underlying device is still there, so delete it. This
1559 * happens when the ofproto is being destroyed, since the caller
1560 * assumes that removal of attached ports will happen as part of
1562 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1565 sset_find_and_delete(&ofproto->ports, devname);
1566 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1567 ofproto->need_revalidate = REV_RECONFIGURE;
1568 bundle_remove(port_);
1569 set_cfm(port_, NULL);
1570 if (ofproto->sflow) {
1571 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1574 ofport_clear_priorities(port);
1575 hmap_destroy(&port->priorities);
1579 port_modified(struct ofport *port_)
1581 struct ofport_dpif *port = ofport_dpif_cast(port_);
1583 if (port->bundle && port->bundle->bond) {
1584 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1589 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1591 struct ofport_dpif *port = ofport_dpif_cast(port_);
1592 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1593 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1595 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1596 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1597 OFPUTIL_PC_NO_PACKET_IN)) {
1598 ofproto->need_revalidate = REV_RECONFIGURE;
1600 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1601 bundle_update(port->bundle);
1607 set_sflow(struct ofproto *ofproto_,
1608 const struct ofproto_sflow_options *sflow_options)
1610 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1611 struct dpif_sflow *ds = ofproto->sflow;
1613 if (sflow_options) {
1615 struct ofport_dpif *ofport;
1617 ds = ofproto->sflow = dpif_sflow_create();
1618 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1619 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1621 ofproto->need_revalidate = REV_RECONFIGURE;
1623 dpif_sflow_set_options(ds, sflow_options);
1626 dpif_sflow_destroy(ds);
1627 ofproto->need_revalidate = REV_RECONFIGURE;
1628 ofproto->sflow = NULL;
1635 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1637 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1644 struct ofproto_dpif *ofproto;
1646 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1647 ofproto->need_revalidate = REV_RECONFIGURE;
1648 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1651 if (cfm_configure(ofport->cfm, s)) {
1657 cfm_destroy(ofport->cfm);
1663 get_cfm_fault(const struct ofport *ofport_)
1665 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1667 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1671 get_cfm_opup(const struct ofport *ofport_)
1673 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1675 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1679 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1682 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1685 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1693 get_cfm_health(const struct ofport *ofport_)
1695 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1697 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1700 /* Spanning Tree. */
1703 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1705 struct ofproto_dpif *ofproto = ofproto_;
1706 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1707 struct ofport_dpif *ofport;
1709 ofport = stp_port_get_aux(sp);
1711 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1712 ofproto->up.name, port_num);
1714 struct eth_header *eth = pkt->l2;
1716 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1717 if (eth_addr_is_zero(eth->eth_src)) {
1718 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1719 "with unknown MAC", ofproto->up.name, port_num);
1721 send_packet(ofport, pkt);
1727 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1729 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1731 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1733 /* Only revalidate flows if the configuration changed. */
1734 if (!s != !ofproto->stp) {
1735 ofproto->need_revalidate = REV_RECONFIGURE;
1739 if (!ofproto->stp) {
1740 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1741 send_bpdu_cb, ofproto);
1742 ofproto->stp_last_tick = time_msec();
1745 stp_set_bridge_id(ofproto->stp, s->system_id);
1746 stp_set_bridge_priority(ofproto->stp, s->priority);
1747 stp_set_hello_time(ofproto->stp, s->hello_time);
1748 stp_set_max_age(ofproto->stp, s->max_age);
1749 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1751 struct ofport *ofport;
1753 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1754 set_stp_port(ofport, NULL);
1757 stp_destroy(ofproto->stp);
1758 ofproto->stp = NULL;
1765 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1771 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1772 s->designated_root = stp_get_designated_root(ofproto->stp);
1773 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1782 update_stp_port_state(struct ofport_dpif *ofport)
1784 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1785 enum stp_state state;
1787 /* Figure out new state. */
1788 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1792 if (ofport->stp_state != state) {
1793 enum ofputil_port_state of_state;
1796 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1797 netdev_get_name(ofport->up.netdev),
1798 stp_state_name(ofport->stp_state),
1799 stp_state_name(state));
1800 if (stp_learn_in_state(ofport->stp_state)
1801 != stp_learn_in_state(state)) {
1802 /* xxx Learning action flows should also be flushed. */
1803 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1805 fwd_change = stp_forward_in_state(ofport->stp_state)
1806 != stp_forward_in_state(state);
1808 ofproto->need_revalidate = REV_STP;
1809 ofport->stp_state = state;
1810 ofport->stp_state_entered = time_msec();
1812 if (fwd_change && ofport->bundle) {
1813 bundle_update(ofport->bundle);
1816 /* Update the STP state bits in the OpenFlow port description. */
1817 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1818 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1819 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1820 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1821 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1823 ofproto_port_set_state(&ofport->up, of_state);
1827 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1828 * caller is responsible for assigning STP port numbers and ensuring
1829 * there are no duplicates. */
1831 set_stp_port(struct ofport *ofport_,
1832 const struct ofproto_port_stp_settings *s)
1834 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1835 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1836 struct stp_port *sp = ofport->stp_port;
1838 if (!s || !s->enable) {
1840 ofport->stp_port = NULL;
1841 stp_port_disable(sp);
1842 update_stp_port_state(ofport);
1845 } else if (sp && stp_port_no(sp) != s->port_num
1846 && ofport == stp_port_get_aux(sp)) {
1847 /* The port-id changed, so disable the old one if it's not
1848 * already in use by another port. */
1849 stp_port_disable(sp);
1852 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1853 stp_port_enable(sp);
1855 stp_port_set_aux(sp, ofport);
1856 stp_port_set_priority(sp, s->priority);
1857 stp_port_set_path_cost(sp, s->path_cost);
1859 update_stp_port_state(ofport);
1865 get_stp_port_status(struct ofport *ofport_,
1866 struct ofproto_port_stp_status *s)
1868 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1869 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1870 struct stp_port *sp = ofport->stp_port;
1872 if (!ofproto->stp || !sp) {
1878 s->port_id = stp_port_get_id(sp);
1879 s->state = stp_port_get_state(sp);
1880 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1881 s->role = stp_port_get_role(sp);
1882 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1888 stp_run(struct ofproto_dpif *ofproto)
1891 long long int now = time_msec();
1892 long long int elapsed = now - ofproto->stp_last_tick;
1893 struct stp_port *sp;
1896 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1897 ofproto->stp_last_tick = now;
1899 while (stp_get_changed_port(ofproto->stp, &sp)) {
1900 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1903 update_stp_port_state(ofport);
1907 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1908 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1914 stp_wait(struct ofproto_dpif *ofproto)
1917 poll_timer_wait(1000);
1921 /* Returns true if STP should process 'flow'. */
1923 stp_should_process_flow(const struct flow *flow)
1925 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1929 stp_process_packet(const struct ofport_dpif *ofport,
1930 const struct ofpbuf *packet)
1932 struct ofpbuf payload = *packet;
1933 struct eth_header *eth = payload.data;
1934 struct stp_port *sp = ofport->stp_port;
1936 /* Sink packets on ports that have STP disabled when the bridge has
1938 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1942 /* Trim off padding on payload. */
1943 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1944 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1947 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1948 stp_received_bpdu(sp, payload.data, payload.size);
1952 static struct priority_to_dscp *
1953 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1955 struct priority_to_dscp *pdscp;
1958 hash = hash_int(priority, 0);
1959 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1960 if (pdscp->priority == priority) {
1968 ofport_clear_priorities(struct ofport_dpif *ofport)
1970 struct priority_to_dscp *pdscp, *next;
1972 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1973 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1979 set_queues(struct ofport *ofport_,
1980 const struct ofproto_port_queue *qdscp_list,
1983 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1985 struct hmap new = HMAP_INITIALIZER(&new);
1988 for (i = 0; i < n_qdscp; i++) {
1989 struct priority_to_dscp *pdscp;
1993 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1994 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
1999 pdscp = get_priority(ofport, priority);
2001 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2003 pdscp = xmalloc(sizeof *pdscp);
2004 pdscp->priority = priority;
2006 ofproto->need_revalidate = REV_RECONFIGURE;
2009 if (pdscp->dscp != dscp) {
2011 ofproto->need_revalidate = REV_RECONFIGURE;
2014 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2017 if (!hmap_is_empty(&ofport->priorities)) {
2018 ofport_clear_priorities(ofport);
2019 ofproto->need_revalidate = REV_RECONFIGURE;
2022 hmap_swap(&new, &ofport->priorities);
2030 /* Expires all MAC learning entries associated with 'bundle' and forces its
2031 * ofproto to revalidate every flow.
2033 * Normally MAC learning entries are removed only from the ofproto associated
2034 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2035 * are removed from every ofproto. When patch ports and SLB bonds are in use
2036 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2037 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2038 * with the host from which it migrated. */
2040 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2042 struct ofproto_dpif *ofproto = bundle->ofproto;
2043 struct mac_learning *ml = ofproto->ml;
2044 struct mac_entry *mac, *next_mac;
2046 ofproto->need_revalidate = REV_RECONFIGURE;
2047 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2048 if (mac->port.p == bundle) {
2050 struct ofproto_dpif *o;
2052 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2054 struct mac_entry *e;
2056 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2059 tag_set_add(&o->revalidate_set, e->tag);
2060 mac_learning_expire(o->ml, e);
2066 mac_learning_expire(ml, mac);
2071 static struct ofbundle *
2072 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2074 struct ofbundle *bundle;
2076 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2077 &ofproto->bundles) {
2078 if (bundle->aux == aux) {
2085 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2086 * ones that are found to 'bundles'. */
2088 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2089 void **auxes, size_t n_auxes,
2090 struct hmapx *bundles)
2094 hmapx_init(bundles);
2095 for (i = 0; i < n_auxes; i++) {
2096 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2098 hmapx_add(bundles, bundle);
2104 bundle_update(struct ofbundle *bundle)
2106 struct ofport_dpif *port;
2108 bundle->floodable = true;
2109 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2110 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2111 || !stp_forward_in_state(port->stp_state)) {
2112 bundle->floodable = false;
2119 bundle_del_port(struct ofport_dpif *port)
2121 struct ofbundle *bundle = port->bundle;
2123 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2125 list_remove(&port->bundle_node);
2126 port->bundle = NULL;
2129 lacp_slave_unregister(bundle->lacp, port);
2132 bond_slave_unregister(bundle->bond, port);
2135 bundle_update(bundle);
2139 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2140 struct lacp_slave_settings *lacp,
2141 uint32_t bond_stable_id)
2143 struct ofport_dpif *port;
2145 port = get_ofp_port(bundle->ofproto, ofp_port);
2150 if (port->bundle != bundle) {
2151 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2153 bundle_del_port(port);
2156 port->bundle = bundle;
2157 list_push_back(&bundle->ports, &port->bundle_node);
2158 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2159 || !stp_forward_in_state(port->stp_state)) {
2160 bundle->floodable = false;
2164 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2165 lacp_slave_register(bundle->lacp, port, lacp);
2168 port->bond_stable_id = bond_stable_id;
2174 bundle_destroy(struct ofbundle *bundle)
2176 struct ofproto_dpif *ofproto;
2177 struct ofport_dpif *port, *next_port;
2184 ofproto = bundle->ofproto;
2185 for (i = 0; i < MAX_MIRRORS; i++) {
2186 struct ofmirror *m = ofproto->mirrors[i];
2188 if (m->out == bundle) {
2190 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2191 || hmapx_find_and_delete(&m->dsts, bundle)) {
2192 ofproto->need_revalidate = REV_RECONFIGURE;
2197 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2198 bundle_del_port(port);
2201 bundle_flush_macs(bundle, true);
2202 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2204 free(bundle->trunks);
2205 lacp_destroy(bundle->lacp);
2206 bond_destroy(bundle->bond);
2211 bundle_set(struct ofproto *ofproto_, void *aux,
2212 const struct ofproto_bundle_settings *s)
2214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2215 bool need_flush = false;
2216 struct ofport_dpif *port;
2217 struct ofbundle *bundle;
2218 unsigned long *trunks;
2224 bundle_destroy(bundle_lookup(ofproto, aux));
2228 assert(s->n_slaves == 1 || s->bond != NULL);
2229 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2231 bundle = bundle_lookup(ofproto, aux);
2233 bundle = xmalloc(sizeof *bundle);
2235 bundle->ofproto = ofproto;
2236 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2237 hash_pointer(aux, 0));
2239 bundle->name = NULL;
2241 list_init(&bundle->ports);
2242 bundle->vlan_mode = PORT_VLAN_TRUNK;
2244 bundle->trunks = NULL;
2245 bundle->use_priority_tags = s->use_priority_tags;
2246 bundle->lacp = NULL;
2247 bundle->bond = NULL;
2249 bundle->floodable = true;
2251 bundle->src_mirrors = 0;
2252 bundle->dst_mirrors = 0;
2253 bundle->mirror_out = 0;
2256 if (!bundle->name || strcmp(s->name, bundle->name)) {
2258 bundle->name = xstrdup(s->name);
2263 if (!bundle->lacp) {
2264 ofproto->need_revalidate = REV_RECONFIGURE;
2265 bundle->lacp = lacp_create();
2267 lacp_configure(bundle->lacp, s->lacp);
2269 lacp_destroy(bundle->lacp);
2270 bundle->lacp = NULL;
2273 /* Update set of ports. */
2275 for (i = 0; i < s->n_slaves; i++) {
2276 if (!bundle_add_port(bundle, s->slaves[i],
2277 s->lacp ? &s->lacp_slaves[i] : NULL,
2278 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2282 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2283 struct ofport_dpif *next_port;
2285 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2286 for (i = 0; i < s->n_slaves; i++) {
2287 if (s->slaves[i] == port->up.ofp_port) {
2292 bundle_del_port(port);
2296 assert(list_size(&bundle->ports) <= s->n_slaves);
2298 if (list_is_empty(&bundle->ports)) {
2299 bundle_destroy(bundle);
2303 /* Set VLAN tagging mode */
2304 if (s->vlan_mode != bundle->vlan_mode
2305 || s->use_priority_tags != bundle->use_priority_tags) {
2306 bundle->vlan_mode = s->vlan_mode;
2307 bundle->use_priority_tags = s->use_priority_tags;
2312 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2313 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2315 if (vlan != bundle->vlan) {
2316 bundle->vlan = vlan;
2320 /* Get trunked VLANs. */
2321 switch (s->vlan_mode) {
2322 case PORT_VLAN_ACCESS:
2326 case PORT_VLAN_TRUNK:
2327 trunks = CONST_CAST(unsigned long *, s->trunks);
2330 case PORT_VLAN_NATIVE_UNTAGGED:
2331 case PORT_VLAN_NATIVE_TAGGED:
2332 if (vlan != 0 && (!s->trunks
2333 || !bitmap_is_set(s->trunks, vlan)
2334 || bitmap_is_set(s->trunks, 0))) {
2335 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2337 trunks = bitmap_clone(s->trunks, 4096);
2339 trunks = bitmap_allocate1(4096);
2341 bitmap_set1(trunks, vlan);
2342 bitmap_set0(trunks, 0);
2344 trunks = CONST_CAST(unsigned long *, s->trunks);
2351 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2352 free(bundle->trunks);
2353 if (trunks == s->trunks) {
2354 bundle->trunks = vlan_bitmap_clone(trunks);
2356 bundle->trunks = trunks;
2361 if (trunks != s->trunks) {
2366 if (!list_is_short(&bundle->ports)) {
2367 bundle->ofproto->has_bonded_bundles = true;
2369 if (bond_reconfigure(bundle->bond, s->bond)) {
2370 ofproto->need_revalidate = REV_RECONFIGURE;
2373 bundle->bond = bond_create(s->bond);
2374 ofproto->need_revalidate = REV_RECONFIGURE;
2377 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2378 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2382 bond_destroy(bundle->bond);
2383 bundle->bond = NULL;
2386 /* If we changed something that would affect MAC learning, un-learn
2387 * everything on this port and force flow revalidation. */
2389 bundle_flush_macs(bundle, false);
2396 bundle_remove(struct ofport *port_)
2398 struct ofport_dpif *port = ofport_dpif_cast(port_);
2399 struct ofbundle *bundle = port->bundle;
2402 bundle_del_port(port);
2403 if (list_is_empty(&bundle->ports)) {
2404 bundle_destroy(bundle);
2405 } else if (list_is_short(&bundle->ports)) {
2406 bond_destroy(bundle->bond);
2407 bundle->bond = NULL;
2413 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2415 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2416 struct ofport_dpif *port = port_;
2417 uint8_t ea[ETH_ADDR_LEN];
2420 error = netdev_get_etheraddr(port->up.netdev, ea);
2422 struct ofpbuf packet;
2425 ofpbuf_init(&packet, 0);
2426 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2428 memcpy(packet_pdu, pdu, pdu_size);
2430 send_packet(port, &packet);
2431 ofpbuf_uninit(&packet);
2433 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2434 "%s (%s)", port->bundle->name,
2435 netdev_get_name(port->up.netdev), strerror(error));
2440 bundle_send_learning_packets(struct ofbundle *bundle)
2442 struct ofproto_dpif *ofproto = bundle->ofproto;
2443 int error, n_packets, n_errors;
2444 struct mac_entry *e;
2446 error = n_packets = n_errors = 0;
2447 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2448 if (e->port.p != bundle) {
2449 struct ofpbuf *learning_packet;
2450 struct ofport_dpif *port;
2454 /* The assignment to "port" is unnecessary but makes "grep"ing for
2455 * struct ofport_dpif more effective. */
2456 learning_packet = bond_compose_learning_packet(bundle->bond,
2460 ret = send_packet(port, learning_packet);
2461 ofpbuf_delete(learning_packet);
2471 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2472 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2473 "packets, last error was: %s",
2474 bundle->name, n_errors, n_packets, strerror(error));
2476 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2477 bundle->name, n_packets);
2482 bundle_run(struct ofbundle *bundle)
2485 lacp_run(bundle->lacp, send_pdu_cb);
2488 struct ofport_dpif *port;
2490 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2491 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2494 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2495 lacp_status(bundle->lacp));
2496 if (bond_should_send_learning_packets(bundle->bond)) {
2497 bundle_send_learning_packets(bundle);
2503 bundle_wait(struct ofbundle *bundle)
2506 lacp_wait(bundle->lacp);
2509 bond_wait(bundle->bond);
2516 mirror_scan(struct ofproto_dpif *ofproto)
2520 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2521 if (!ofproto->mirrors[idx]) {
2528 static struct ofmirror *
2529 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2533 for (i = 0; i < MAX_MIRRORS; i++) {
2534 struct ofmirror *mirror = ofproto->mirrors[i];
2535 if (mirror && mirror->aux == aux) {
2543 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2545 mirror_update_dups(struct ofproto_dpif *ofproto)
2549 for (i = 0; i < MAX_MIRRORS; i++) {
2550 struct ofmirror *m = ofproto->mirrors[i];
2553 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2557 for (i = 0; i < MAX_MIRRORS; i++) {
2558 struct ofmirror *m1 = ofproto->mirrors[i];
2565 for (j = i + 1; j < MAX_MIRRORS; j++) {
2566 struct ofmirror *m2 = ofproto->mirrors[j];
2568 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2569 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2570 m2->dup_mirrors |= m1->dup_mirrors;
2577 mirror_set(struct ofproto *ofproto_, void *aux,
2578 const struct ofproto_mirror_settings *s)
2580 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2581 mirror_mask_t mirror_bit;
2582 struct ofbundle *bundle;
2583 struct ofmirror *mirror;
2584 struct ofbundle *out;
2585 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2586 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2589 mirror = mirror_lookup(ofproto, aux);
2591 mirror_destroy(mirror);
2597 idx = mirror_scan(ofproto);
2599 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2601 ofproto->up.name, MAX_MIRRORS, s->name);
2605 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2606 mirror->ofproto = ofproto;
2609 mirror->out_vlan = -1;
2610 mirror->name = NULL;
2613 if (!mirror->name || strcmp(s->name, mirror->name)) {
2615 mirror->name = xstrdup(s->name);
2618 /* Get the new configuration. */
2619 if (s->out_bundle) {
2620 out = bundle_lookup(ofproto, s->out_bundle);
2622 mirror_destroy(mirror);
2628 out_vlan = s->out_vlan;
2630 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2631 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2633 /* If the configuration has not changed, do nothing. */
2634 if (hmapx_equals(&srcs, &mirror->srcs)
2635 && hmapx_equals(&dsts, &mirror->dsts)
2636 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2637 && mirror->out == out
2638 && mirror->out_vlan == out_vlan)
2640 hmapx_destroy(&srcs);
2641 hmapx_destroy(&dsts);
2645 hmapx_swap(&srcs, &mirror->srcs);
2646 hmapx_destroy(&srcs);
2648 hmapx_swap(&dsts, &mirror->dsts);
2649 hmapx_destroy(&dsts);
2651 free(mirror->vlans);
2652 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2655 mirror->out_vlan = out_vlan;
2657 /* Update bundles. */
2658 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2659 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2660 if (hmapx_contains(&mirror->srcs, bundle)) {
2661 bundle->src_mirrors |= mirror_bit;
2663 bundle->src_mirrors &= ~mirror_bit;
2666 if (hmapx_contains(&mirror->dsts, bundle)) {
2667 bundle->dst_mirrors |= mirror_bit;
2669 bundle->dst_mirrors &= ~mirror_bit;
2672 if (mirror->out == bundle) {
2673 bundle->mirror_out |= mirror_bit;
2675 bundle->mirror_out &= ~mirror_bit;
2679 ofproto->need_revalidate = REV_RECONFIGURE;
2680 ofproto->has_mirrors = true;
2681 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2682 mirror_update_dups(ofproto);
2688 mirror_destroy(struct ofmirror *mirror)
2690 struct ofproto_dpif *ofproto;
2691 mirror_mask_t mirror_bit;
2692 struct ofbundle *bundle;
2699 ofproto = mirror->ofproto;
2700 ofproto->need_revalidate = REV_RECONFIGURE;
2701 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2703 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2704 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2705 bundle->src_mirrors &= ~mirror_bit;
2706 bundle->dst_mirrors &= ~mirror_bit;
2707 bundle->mirror_out &= ~mirror_bit;
2710 hmapx_destroy(&mirror->srcs);
2711 hmapx_destroy(&mirror->dsts);
2712 free(mirror->vlans);
2714 ofproto->mirrors[mirror->idx] = NULL;
2718 mirror_update_dups(ofproto);
2720 ofproto->has_mirrors = false;
2721 for (i = 0; i < MAX_MIRRORS; i++) {
2722 if (ofproto->mirrors[i]) {
2723 ofproto->has_mirrors = true;
2730 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2731 uint64_t *packets, uint64_t *bytes)
2733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2734 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2737 *packets = *bytes = UINT64_MAX;
2741 *packets = mirror->packet_count;
2742 *bytes = mirror->byte_count;
2748 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2750 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2751 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2752 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2758 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2761 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2762 return bundle && bundle->mirror_out != 0;
2766 forward_bpdu_changed(struct ofproto *ofproto_)
2768 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2769 ofproto->need_revalidate = REV_RECONFIGURE;
2773 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2776 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2777 mac_learning_set_idle_time(ofproto->ml, idle_time);
2778 mac_learning_set_max_entries(ofproto->ml, max_entries);
2783 static struct ofport_dpif *
2784 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2786 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2787 return ofport ? ofport_dpif_cast(ofport) : NULL;
2790 static struct ofport_dpif *
2791 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2793 return get_ofp_port(ofproto, odp_port_to_ofp_port(ofproto, odp_port));
2797 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2798 struct ofproto_port *ofproto_port,
2799 struct dpif_port *dpif_port)
2801 ofproto_port->name = dpif_port->name;
2802 ofproto_port->type = dpif_port->type;
2803 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2807 port_run_fast(struct ofport_dpif *ofport)
2809 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2810 struct ofpbuf packet;
2812 ofpbuf_init(&packet, 0);
2813 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2814 send_packet(ofport, &packet);
2815 ofpbuf_uninit(&packet);
2820 port_run(struct ofport_dpif *ofport)
2822 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2823 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2824 bool enable = netdev_get_carrier(ofport->up.netdev);
2826 ofport->carrier_seq = carrier_seq;
2828 port_run_fast(ofport);
2830 int cfm_opup = cfm_get_opup(ofport->cfm);
2832 cfm_run(ofport->cfm);
2833 enable = enable && !cfm_get_fault(ofport->cfm);
2835 if (cfm_opup >= 0) {
2836 enable = enable && cfm_opup;
2840 if (ofport->bundle) {
2841 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2842 if (carrier_changed) {
2843 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2847 if (ofport->may_enable != enable) {
2848 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2850 if (ofproto->has_bundle_action) {
2851 ofproto->need_revalidate = REV_PORT_TOGGLED;
2855 ofport->may_enable = enable;
2859 port_wait(struct ofport_dpif *ofport)
2862 cfm_wait(ofport->cfm);
2867 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2868 struct ofproto_port *ofproto_port)
2870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2871 struct dpif_port dpif_port;
2874 if (!sset_contains(&ofproto->ports, devname)) {
2877 error = dpif_port_query_by_name(ofproto->backer->dpif,
2878 devname, &dpif_port);
2880 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2886 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2889 uint32_t odp_port = UINT32_MAX;
2892 error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port);
2894 sset_add(&ofproto->ports, netdev_get_name(netdev));
2900 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2902 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2903 uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port);
2906 if (odp_port != OFPP_NONE) {
2907 error = dpif_port_del(ofproto->backer->dpif, odp_port);
2910 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2912 /* The caller is going to close ofport->up.netdev. If this is a
2913 * bonded port, then the bond is using that netdev, so remove it
2914 * from the bond. The client will need to reconfigure everything
2915 * after deleting ports, so then the slave will get re-added. */
2916 bundle_remove(&ofport->up);
2923 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2925 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2928 error = netdev_get_stats(ofport->up.netdev, stats);
2930 if (!error && ofport->odp_port == OVSP_LOCAL) {
2931 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2933 /* ofproto->stats.tx_packets represents packets that we created
2934 * internally and sent to some port (e.g. packets sent with
2935 * send_packet()). Account for them as if they had come from
2936 * OFPP_LOCAL and got forwarded. */
2938 if (stats->rx_packets != UINT64_MAX) {
2939 stats->rx_packets += ofproto->stats.tx_packets;
2942 if (stats->rx_bytes != UINT64_MAX) {
2943 stats->rx_bytes += ofproto->stats.tx_bytes;
2946 /* ofproto->stats.rx_packets represents packets that were received on
2947 * some port and we processed internally and dropped (e.g. STP).
2948 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2950 if (stats->tx_packets != UINT64_MAX) {
2951 stats->tx_packets += ofproto->stats.rx_packets;
2954 if (stats->tx_bytes != UINT64_MAX) {
2955 stats->tx_bytes += ofproto->stats.rx_bytes;
2962 /* Account packets for LOCAL port. */
2964 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2965 size_t tx_size, size_t rx_size)
2967 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2970 ofproto->stats.rx_packets++;
2971 ofproto->stats.rx_bytes += rx_size;
2974 ofproto->stats.tx_packets++;
2975 ofproto->stats.tx_bytes += tx_size;
2979 struct port_dump_state {
2985 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2987 struct port_dump_state *state;
2989 *statep = state = xmalloc(sizeof *state);
2996 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2997 struct ofproto_port *port)
2999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3000 struct port_dump_state *state = state_;
3001 struct sset_node *node;
3003 while ((node = sset_at_position(&ofproto->ports, &state->bucket,
3007 error = port_query_by_name(ofproto_, node->name, port);
3008 if (error != ENODEV) {
3017 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3019 struct port_dump_state *state = state_;
3026 port_poll(const struct ofproto *ofproto_, char **devnamep)
3028 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3030 if (ofproto->port_poll_errno) {
3031 int error = ofproto->port_poll_errno;
3032 ofproto->port_poll_errno = 0;
3036 if (sset_is_empty(&ofproto->port_poll_set)) {
3040 *devnamep = sset_pop(&ofproto->port_poll_set);
3045 port_poll_wait(const struct ofproto *ofproto_)
3047 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3048 dpif_port_poll_wait(ofproto->backer->dpif);
3052 port_is_lacp_current(const struct ofport *ofport_)
3054 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3055 return (ofport->bundle && ofport->bundle->lacp
3056 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3060 /* Upcall handling. */
3062 /* Flow miss batching.
3064 * Some dpifs implement operations faster when you hand them off in a batch.
3065 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3066 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3067 * more packets, plus possibly installing the flow in the dpif.
3069 * So far we only batch the operations that affect flow setup time the most.
3070 * It's possible to batch more than that, but the benefit might be minimal. */
3072 struct hmap_node hmap_node;
3073 struct ofproto_dpif *ofproto;
3075 enum odp_key_fitness key_fitness;
3076 const struct nlattr *key;
3078 ovs_be16 initial_tci;
3079 struct list packets;
3080 enum dpif_upcall_type upcall_type;
3081 uint32_t odp_in_port;
3084 struct flow_miss_op {
3085 struct dpif_op dpif_op;
3086 struct subfacet *subfacet; /* Subfacet */
3087 void *garbage; /* Pointer to pass to free(), NULL if none. */
3088 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3091 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3092 * OpenFlow controller as necessary according to their individual
3093 * configurations. */
3095 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3096 const struct flow *flow)
3098 struct ofputil_packet_in pin;
3100 pin.packet = packet->data;
3101 pin.packet_len = packet->size;
3102 pin.reason = OFPR_NO_MATCH;
3103 pin.controller_id = 0;
3108 pin.send_len = 0; /* not used for flow table misses */
3110 flow_get_metadata(flow, &pin.fmd);
3112 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3115 static enum slow_path_reason
3116 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3117 const struct ofpbuf *packet)
3119 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
3125 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3127 cfm_process_heartbeat(ofport->cfm, packet);
3130 } else if (ofport->bundle && ofport->bundle->lacp
3131 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3133 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3136 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3138 stp_process_packet(ofport, packet);
3145 static struct flow_miss *
3146 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3148 struct flow_miss *miss;
3150 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3151 if (flow_equal(&miss->flow, flow)) {
3159 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3160 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3161 * 'miss' is associated with a subfacet the caller must also initialize the
3162 * returned op->subfacet, and if anything needs to be freed after processing
3163 * the op, the caller must initialize op->garbage also. */
3165 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3166 struct flow_miss_op *op)
3168 if (miss->flow.vlan_tci != miss->initial_tci) {
3169 /* This packet was received on a VLAN splinter port. We
3170 * added a VLAN to the packet to make the packet resemble
3171 * the flow, but the actions were composed assuming that
3172 * the packet contained no VLAN. So, we must remove the
3173 * VLAN header from the packet before trying to execute the
3175 eth_pop_vlan(packet);
3178 op->subfacet = NULL;
3180 op->dpif_op.type = DPIF_OP_EXECUTE;
3181 op->dpif_op.u.execute.key = miss->key;
3182 op->dpif_op.u.execute.key_len = miss->key_len;
3183 op->dpif_op.u.execute.packet = packet;
3186 /* Helper for handle_flow_miss_without_facet() and
3187 * handle_flow_miss_with_facet(). */
3189 handle_flow_miss_common(struct rule_dpif *rule,
3190 struct ofpbuf *packet, const struct flow *flow)
3192 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3194 ofproto->n_matches++;
3196 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3198 * Extra-special case for fail-open mode.
3200 * We are in fail-open mode and the packet matched the fail-open
3201 * rule, but we are connected to a controller too. We should send
3202 * the packet up to the controller in the hope that it will try to
3203 * set up a flow and thereby allow us to exit fail-open.
3205 * See the top-level comment in fail-open.c for more information.
3207 send_packet_in_miss(ofproto, packet, flow);
3211 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3212 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3213 * installing a datapath flow. The answer is usually "yes" (a return value of
3214 * true). However, for short flows the cost of bookkeeping is much higher than
3215 * the benefits, so when the datapath holds a large number of flows we impose
3216 * some heuristics to decide which flows are likely to be worth tracking. */
3218 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3219 struct flow_miss *miss, uint32_t hash)
3221 if (!ofproto->governor) {
3224 n_subfacets = hmap_count(&ofproto->subfacets);
3225 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3229 ofproto->governor = governor_create(ofproto->up.name);
3232 return governor_should_install_flow(ofproto->governor, hash,
3233 list_size(&miss->packets));
3236 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3237 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3238 * increment '*n_ops'. */
3240 handle_flow_miss_without_facet(struct flow_miss *miss,
3241 struct rule_dpif *rule,
3242 struct flow_miss_op *ops, size_t *n_ops)
3244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3245 long long int now = time_msec();
3246 struct action_xlate_ctx ctx;
3247 struct ofpbuf *packet;
3249 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3250 struct flow_miss_op *op = &ops[*n_ops];
3251 struct dpif_flow_stats stats;
3252 struct ofpbuf odp_actions;
3254 COVERAGE_INC(facet_suppress);
3256 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3258 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3259 rule_credit_stats(rule, &stats);
3261 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3263 ctx.resubmit_stats = &stats;
3264 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3267 if (odp_actions.size) {
3268 struct dpif_execute *execute = &op->dpif_op.u.execute;
3270 init_flow_miss_execute_op(miss, packet, op);
3271 execute->actions = odp_actions.data;
3272 execute->actions_len = odp_actions.size;
3273 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3277 ofpbuf_uninit(&odp_actions);
3282 /* Handles 'miss', which matches 'facet'. May add any required datapath
3283 * operations to 'ops', incrementing '*n_ops' for each new op.
3285 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3286 * This is really important only for new facets: if we just called time_msec()
3287 * here, then the new subfacet or its packets could look (occasionally) as
3288 * though it was used some time after the facet was used. That can make a
3289 * one-packet flow look like it has a nonzero duration, which looks odd in
3290 * e.g. NetFlow statistics. */
3292 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3294 struct flow_miss_op *ops, size_t *n_ops)
3296 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3297 enum subfacet_path want_path;
3298 struct subfacet *subfacet;
3299 struct ofpbuf *packet;
3301 subfacet = subfacet_create(facet, miss, now);
3303 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3304 struct flow_miss_op *op = &ops[*n_ops];
3305 struct dpif_flow_stats stats;
3306 struct ofpbuf odp_actions;
3308 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3310 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3311 if (!subfacet->actions || subfacet->slow) {
3312 subfacet_make_actions(subfacet, packet, &odp_actions);
3315 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3316 subfacet_update_stats(subfacet, &stats);
3318 if (subfacet->actions_len) {
3319 struct dpif_execute *execute = &op->dpif_op.u.execute;
3321 init_flow_miss_execute_op(miss, packet, op);
3322 op->subfacet = subfacet;
3323 if (!subfacet->slow) {
3324 execute->actions = subfacet->actions;
3325 execute->actions_len = subfacet->actions_len;
3326 ofpbuf_uninit(&odp_actions);
3328 execute->actions = odp_actions.data;
3329 execute->actions_len = odp_actions.size;
3330 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3335 ofpbuf_uninit(&odp_actions);
3339 want_path = subfacet_want_path(subfacet->slow);
3340 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3341 struct flow_miss_op *op = &ops[(*n_ops)++];
3342 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3344 op->subfacet = subfacet;
3346 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3347 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3348 put->key = miss->key;
3349 put->key_len = miss->key_len;
3350 if (want_path == SF_FAST_PATH) {
3351 put->actions = subfacet->actions;
3352 put->actions_len = subfacet->actions_len;
3354 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3355 op->stub, sizeof op->stub,
3356 &put->actions, &put->actions_len);
3362 /* Handles flow miss 'miss'. May add any required datapath operations
3363 * to 'ops', incrementing '*n_ops' for each new op. */
3365 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3368 struct ofproto_dpif *ofproto = miss->ofproto;
3369 struct facet *facet;
3373 /* The caller must ensure that miss->hmap_node.hash contains
3374 * flow_hash(miss->flow, 0). */
3375 hash = miss->hmap_node.hash;
3377 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3379 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3381 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3382 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3386 facet = facet_create(rule, &miss->flow, hash);
3391 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3394 /* This function does post-processing on data returned from
3395 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the
3396 * rest of the upcall processing logic. In particular, if the extracted
3397 * in_port is a VLAN splinter port, it replaces flow->in_port by the "real"
3398 * port, sets flow->vlan_tci correctly for the VLAN of the VLAN splinter
3399 * port, and pushes a VLAN header onto 'packet' (if it is nonnull). The
3400 * caller must have called odp_flow_key_to_flow() and supply 'fitness' and
3401 * 'flow' from its output. The 'flow' argument must have had the "in_port"
3402 * member converted to the OpenFlow number.
3404 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
3405 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
3406 * (This differs from the value returned in flow->vlan_tci only for packets
3407 * received on VLAN splinters.) */
3408 static enum odp_key_fitness
3409 ofproto_dpif_vsp_adjust(const struct ofproto_dpif *ofproto,
3410 enum odp_key_fitness fitness,
3411 struct flow *flow, ovs_be16 *initial_tci,
3412 struct ofpbuf *packet)
3414 if (fitness == ODP_FIT_ERROR) {
3417 *initial_tci = flow->vlan_tci;
3419 if (vsp_adjust_flow(ofproto, flow)) {
3421 /* Make the packet resemble the flow, so that it gets sent to an
3422 * OpenFlow controller properly, so that it looks correct for
3423 * sFlow, and so that flow_extract() will get the correct vlan_tci
3424 * if it is called on 'packet'.
3426 * The allocated space inside 'packet' probably also contains
3427 * 'key', that is, both 'packet' and 'key' are probably part of a
3428 * struct dpif_upcall (see the large comment on that structure
3429 * definition), so pushing data on 'packet' is in general not a
3430 * good idea since it could overwrite 'key' or free it as a side
3431 * effect. However, it's OK in this special case because we know
3432 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3433 * will just overwrite the 4-byte "struct nlattr", which is fine
3434 * since we don't need that header anymore. */
3435 eth_push_vlan(packet, flow->vlan_tci);
3438 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3439 if (fitness == ODP_FIT_PERFECT) {
3440 fitness = ODP_FIT_TOO_MUCH;
3448 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3451 struct dpif_upcall *upcall;
3452 struct flow_miss *miss;
3453 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3454 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3455 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3465 /* Construct the to-do list.
3467 * This just amounts to extracting the flow from each packet and sticking
3468 * the packets that have the same flow in the same "flow_miss" structure so
3469 * that we can process them together. */
3472 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3473 struct flow_miss *miss = &misses[n_misses];
3474 struct flow_miss *existing_miss;
3475 enum odp_key_fitness fitness;
3476 struct ofproto_dpif *ofproto;
3477 struct ofport_dpif *port;
3478 uint32_t odp_in_port;
3482 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3483 port = odp_port_to_ofport(backer, flow.in_port);
3485 /* Received packet on port for which we couldn't associate
3486 * an ofproto. This can happen if a port is removed while
3487 * traffic is being received. Print a rate-limited message
3488 * in case it happens frequently. */
3489 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3493 ofproto = ofproto_dpif_cast(port->up.ofproto);
3494 odp_in_port = flow.in_port;
3495 flow.in_port = port->up.ofp_port;
3497 /* Obtain metadata and check userspace/kernel agreement on flow match,
3498 * then set 'flow''s header pointers. */
3499 miss->key_fitness = ofproto_dpif_vsp_adjust(ofproto, fitness,
3500 &flow, &miss->initial_tci, upcall->packet);
3501 if (miss->key_fitness == ODP_FIT_ERROR) {
3504 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3505 &flow.tunnel, flow.in_port, &miss->flow);
3507 /* Add other packets to a to-do list. */
3508 hash = flow_hash(&miss->flow, 0);
3509 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3510 if (!existing_miss) {
3511 hmap_insert(&todo, &miss->hmap_node, hash);
3512 miss->ofproto = ofproto;
3513 miss->key = upcall->key;
3514 miss->key_len = upcall->key_len;
3515 miss->upcall_type = upcall->type;
3516 miss->odp_in_port = odp_in_port;
3517 list_init(&miss->packets);
3521 miss = existing_miss;
3523 list_push_back(&miss->packets, &upcall->packet->list_node);
3526 /* Process each element in the to-do list, constructing the set of
3527 * operations to batch. */
3529 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3530 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3532 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3534 /* Execute batch. */
3535 for (i = 0; i < n_ops; i++) {
3536 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3538 dpif_operate(backer->dpif, dpif_ops, n_ops);
3540 /* Free memory and update facets. */
3541 for (i = 0; i < n_ops; i++) {
3542 struct flow_miss_op *op = &flow_miss_ops[i];
3544 switch (op->dpif_op.type) {
3545 case DPIF_OP_EXECUTE:
3548 case DPIF_OP_FLOW_PUT:
3549 if (!op->dpif_op.error) {
3550 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3554 case DPIF_OP_FLOW_DEL:
3560 hmap_destroy(&todo);
3563 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3564 classify_upcall(const struct dpif_upcall *upcall)
3566 union user_action_cookie cookie;
3568 /* First look at the upcall type. */
3569 switch (upcall->type) {
3570 case DPIF_UC_ACTION:
3576 case DPIF_N_UC_TYPES:
3578 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3582 /* "action" upcalls need a closer look. */
3583 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3584 switch (cookie.type) {
3585 case USER_ACTION_COOKIE_SFLOW:
3586 return SFLOW_UPCALL;
3588 case USER_ACTION_COOKIE_SLOW_PATH:
3591 case USER_ACTION_COOKIE_UNSPEC:
3593 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3599 handle_sflow_upcall(struct dpif_backer *backer,
3600 const struct dpif_upcall *upcall)
3602 struct ofproto_dpif *ofproto;
3603 union user_action_cookie cookie;
3604 enum odp_key_fitness fitness;
3605 struct ofport_dpif *port;
3606 ovs_be16 initial_tci;
3608 uint32_t odp_in_port;
3610 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3612 port = odp_port_to_ofport(backer, flow.in_port);
3617 ofproto = ofproto_dpif_cast(port->up.ofproto);
3618 if (!ofproto->sflow) {
3622 odp_in_port = flow.in_port;
3623 flow.in_port = port->up.ofp_port;
3624 fitness = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
3625 &initial_tci, upcall->packet);
3626 if (fitness == ODP_FIT_ERROR) {
3630 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3631 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3632 odp_in_port, &cookie);
3636 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3638 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3639 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3640 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3645 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3648 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3649 struct dpif_upcall *upcall = &misses[n_misses];
3650 struct ofpbuf *buf = &miss_bufs[n_misses];
3653 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3654 sizeof miss_buf_stubs[n_misses]);
3655 error = dpif_recv(backer->dpif, upcall, buf);
3661 switch (classify_upcall(upcall)) {
3663 /* Handle it later. */
3668 handle_sflow_upcall(backer, upcall);
3678 /* Handle deferred MISS_UPCALL processing. */
3679 handle_miss_upcalls(backer, misses, n_misses);
3680 for (i = 0; i < n_misses; i++) {
3681 ofpbuf_uninit(&miss_bufs[i]);
3687 /* Flow expiration. */
3689 static int subfacet_max_idle(const struct ofproto_dpif *);
3690 static void update_stats(struct dpif_backer *);
3691 static void rule_expire(struct rule_dpif *);
3692 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3694 /* This function is called periodically by run(). Its job is to collect
3695 * updates for the flows that have been installed into the datapath, most
3696 * importantly when they last were used, and then use that information to
3697 * expire flows that have not been used recently.
3699 * Returns the number of milliseconds after which it should be called again. */
3701 expire(struct dpif_backer *backer)
3703 struct ofproto_dpif *ofproto;
3704 int max_idle = INT32_MAX;
3706 /* Update stats for each flow in the backer. */
3707 update_stats(backer);
3709 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3710 struct rule_dpif *rule, *next_rule;
3711 struct oftable *table;
3714 if (ofproto->backer != backer) {
3718 /* Expire subfacets that have been idle too long. */
3719 dp_max_idle = subfacet_max_idle(ofproto);
3720 expire_subfacets(ofproto, dp_max_idle);
3722 max_idle = MIN(max_idle, dp_max_idle);
3724 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3726 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3727 struct cls_cursor cursor;
3729 cls_cursor_init(&cursor, &table->cls, NULL);
3730 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3735 /* All outstanding data in existing flows has been accounted, so it's a
3736 * good time to do bond rebalancing. */
3737 if (ofproto->has_bonded_bundles) {
3738 struct ofbundle *bundle;
3740 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3742 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3748 return MIN(max_idle, 1000);
3751 /* Updates flow table statistics given that the datapath just reported 'stats'
3752 * as 'subfacet''s statistics. */
3754 update_subfacet_stats(struct subfacet *subfacet,
3755 const struct dpif_flow_stats *stats)
3757 struct facet *facet = subfacet->facet;
3759 if (stats->n_packets >= subfacet->dp_packet_count) {
3760 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3761 facet->packet_count += extra;
3763 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3766 if (stats->n_bytes >= subfacet->dp_byte_count) {
3767 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3769 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3772 subfacet->dp_packet_count = stats->n_packets;
3773 subfacet->dp_byte_count = stats->n_bytes;
3775 facet->tcp_flags |= stats->tcp_flags;
3777 subfacet_update_time(subfacet, stats->used);
3778 if (facet->accounted_bytes < facet->byte_count) {
3780 facet_account(facet);
3781 facet->accounted_bytes = facet->byte_count;
3783 facet_push_stats(facet);
3786 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3787 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3789 delete_unexpected_flow(struct ofproto_dpif *ofproto,
3790 const struct nlattr *key, size_t key_len)
3792 if (!VLOG_DROP_WARN(&rl)) {
3796 odp_flow_key_format(key, key_len, &s);
3797 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
3801 COVERAGE_INC(facet_unexpected);
3802 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
3805 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3807 * This function also pushes statistics updates to rules which each facet
3808 * resubmits into. Generally these statistics will be accurate. However, if a
3809 * facet changes the rule it resubmits into at some time in between
3810 * update_stats() runs, it is possible that statistics accrued to the
3811 * old rule will be incorrectly attributed to the new rule. This could be
3812 * avoided by calling update_stats() whenever rules are created or
3813 * deleted. However, the performance impact of making so many calls to the
3814 * datapath do not justify the benefit of having perfectly accurate statistics.
3817 update_stats(struct dpif_backer *backer)
3819 const struct dpif_flow_stats *stats;
3820 struct dpif_flow_dump dump;
3821 const struct nlattr *key;
3824 dpif_flow_dump_start(&dump, backer->dpif);
3825 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3827 struct subfacet *subfacet;
3828 enum odp_key_fitness fitness;
3829 struct ofproto_dpif *ofproto;
3830 struct ofport_dpif *port;
3833 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3834 if (fitness == ODP_FIT_ERROR) {
3838 port = odp_port_to_ofport(backer, flow.in_port);
3840 /* This flow is for a port for which we couldn't associate an
3841 * ofproto. This can happen if a port is removed while
3842 * traffic is being received. Ignore this flow, since it
3843 * will get timed out. */
3847 ofproto = ofproto_dpif_cast(port->up.ofproto);
3848 flow.in_port = port->up.ofp_port;
3849 key_hash = odp_flow_key_hash(key, key_len);
3851 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
3852 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3854 update_subfacet_stats(subfacet, stats);
3858 /* Stats are updated per-packet. */
3861 case SF_NOT_INSTALLED:
3863 delete_unexpected_flow(ofproto, key, key_len);
3867 dpif_flow_dump_done(&dump);
3870 /* Calculates and returns the number of milliseconds of idle time after which
3871 * subfacets should expire from the datapath. When a subfacet expires, we fold
3872 * its statistics into its facet, and when a facet's last subfacet expires, we
3873 * fold its statistic into its rule. */
3875 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3878 * Idle time histogram.
3880 * Most of the time a switch has a relatively small number of subfacets.
3881 * When this is the case we might as well keep statistics for all of them
3882 * in userspace and to cache them in the kernel datapath for performance as
3885 * As the number of subfacets increases, the memory required to maintain
3886 * statistics about them in userspace and in the kernel becomes
3887 * significant. However, with a large number of subfacets it is likely
3888 * that only a few of them are "heavy hitters" that consume a large amount
3889 * of bandwidth. At this point, only heavy hitters are worth caching in
3890 * the kernel and maintaining in userspaces; other subfacets we can
3893 * The technique used to compute the idle time is to build a histogram with
3894 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3895 * that is installed in the kernel gets dropped in the appropriate bucket.
3896 * After the histogram has been built, we compute the cutoff so that only
3897 * the most-recently-used 1% of subfacets (but at least
3898 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3899 * the most-recently-used bucket of subfacets is kept, so actually an
3900 * arbitrary number of subfacets can be kept in any given expiration run
3901 * (though the next run will delete most of those unless they receive
3904 * This requires a second pass through the subfacets, in addition to the
3905 * pass made by update_stats(), because the former function never looks at
3906 * uninstallable subfacets.
3908 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3909 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3910 int buckets[N_BUCKETS] = { 0 };
3911 int total, subtotal, bucket;
3912 struct subfacet *subfacet;
3916 total = hmap_count(&ofproto->subfacets);
3917 if (total <= ofproto->up.flow_eviction_threshold) {
3918 return N_BUCKETS * BUCKET_WIDTH;
3921 /* Build histogram. */
3923 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3924 long long int idle = now - subfacet->used;
3925 int bucket = (idle <= 0 ? 0
3926 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3927 : (unsigned int) idle / BUCKET_WIDTH);
3931 /* Find the first bucket whose flows should be expired. */
3932 subtotal = bucket = 0;
3934 subtotal += buckets[bucket++];
3935 } while (bucket < N_BUCKETS &&
3936 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3938 if (VLOG_IS_DBG_ENABLED()) {
3942 ds_put_cstr(&s, "keep");
3943 for (i = 0; i < N_BUCKETS; i++) {
3945 ds_put_cstr(&s, ", drop");
3948 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3951 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3955 return bucket * BUCKET_WIDTH;
3959 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3961 /* Cutoff time for most flows. */
3962 long long int normal_cutoff = time_msec() - dp_max_idle;
3964 /* We really want to keep flows for special protocols around, so use a more
3965 * conservative cutoff. */
3966 long long int special_cutoff = time_msec() - 10000;
3968 struct subfacet *subfacet, *next_subfacet;
3969 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3973 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3974 &ofproto->subfacets) {
3975 long long int cutoff;
3977 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3980 if (subfacet->used < cutoff) {
3981 if (subfacet->path != SF_NOT_INSTALLED) {
3982 batch[n_batch++] = subfacet;
3983 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3984 subfacet_destroy_batch(ofproto, batch, n_batch);
3988 subfacet_destroy(subfacet);
3994 subfacet_destroy_batch(ofproto, batch, n_batch);
3998 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3999 * then delete it entirely. */
4001 rule_expire(struct rule_dpif *rule)
4003 struct facet *facet, *next_facet;
4007 if (rule->up.pending) {
4008 /* We'll have to expire it later. */
4012 /* Has 'rule' expired? */
4014 if (rule->up.hard_timeout
4015 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4016 reason = OFPRR_HARD_TIMEOUT;
4017 } else if (rule->up.idle_timeout
4018 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4019 reason = OFPRR_IDLE_TIMEOUT;
4024 COVERAGE_INC(ofproto_dpif_expired);
4026 /* Update stats. (This is a no-op if the rule expired due to an idle
4027 * timeout, because that only happens when the rule has no facets left.) */
4028 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4029 facet_remove(facet);
4032 /* Get rid of the rule. */
4033 ofproto_rule_expire(&rule->up, reason);
4038 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4040 * The caller must already have determined that no facet with an identical
4041 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4042 * the ofproto's classifier table.
4044 * 'hash' must be the return value of flow_hash(flow, 0).
4046 * The facet will initially have no subfacets. The caller should create (at
4047 * least) one subfacet with subfacet_create(). */
4048 static struct facet *
4049 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4052 struct facet *facet;
4054 facet = xzalloc(sizeof *facet);
4055 facet->used = time_msec();
4056 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4057 list_push_back(&rule->facets, &facet->list_node);
4059 facet->flow = *flow;
4060 list_init(&facet->subfacets);
4061 netflow_flow_init(&facet->nf_flow);
4062 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4068 facet_free(struct facet *facet)
4073 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4074 * 'packet', which arrived on 'in_port'.
4076 * Takes ownership of 'packet'. */
4078 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4079 const struct nlattr *odp_actions, size_t actions_len,
4080 struct ofpbuf *packet)
4082 struct odputil_keybuf keybuf;
4086 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4087 odp_flow_key_from_flow(&key, flow,
4088 ofp_port_to_odp_port(ofproto, flow->in_port));
4090 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4091 odp_actions, actions_len, packet);
4093 ofpbuf_delete(packet);
4097 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4099 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4100 * rule's statistics, via subfacet_uninstall().
4102 * - Removes 'facet' from its rule and from ofproto->facets.
4105 facet_remove(struct facet *facet)
4107 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4108 struct subfacet *subfacet, *next_subfacet;
4110 assert(!list_is_empty(&facet->subfacets));
4112 /* First uninstall all of the subfacets to get final statistics. */
4113 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4114 subfacet_uninstall(subfacet);
4117 /* Flush the final stats to the rule.
4119 * This might require us to have at least one subfacet around so that we
4120 * can use its actions for accounting in facet_account(), which is why we
4121 * have uninstalled but not yet destroyed the subfacets. */
4122 facet_flush_stats(facet);
4124 /* Now we're really all done so destroy everything. */
4125 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4126 &facet->subfacets) {
4127 subfacet_destroy__(subfacet);
4129 hmap_remove(&ofproto->facets, &facet->hmap_node);
4130 list_remove(&facet->list_node);
4134 /* Feed information from 'facet' back into the learning table to keep it in
4135 * sync with what is actually flowing through the datapath. */
4137 facet_learn(struct facet *facet)
4139 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4140 struct action_xlate_ctx ctx;
4142 if (!facet->has_learn
4143 && !facet->has_normal
4144 && (!facet->has_fin_timeout
4145 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4149 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4150 facet->flow.vlan_tci,
4151 facet->rule, facet->tcp_flags, NULL);
4152 ctx.may_learn = true;
4153 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4154 facet->rule->up.ofpacts_len);
4158 facet_account(struct facet *facet)
4160 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4161 struct subfacet *subfacet;
4162 const struct nlattr *a;
4167 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4170 n_bytes = facet->byte_count - facet->accounted_bytes;
4172 /* This loop feeds byte counters to bond_account() for rebalancing to use
4173 * as a basis. We also need to track the actual VLAN on which the packet
4174 * is going to be sent to ensure that it matches the one passed to
4175 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4178 * We use the actions from an arbitrary subfacet because they should all
4179 * be equally valid for our purpose. */
4180 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4181 struct subfacet, list_node);
4182 vlan_tci = facet->flow.vlan_tci;
4183 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4184 subfacet->actions, subfacet->actions_len) {
4185 const struct ovs_action_push_vlan *vlan;
4186 struct ofport_dpif *port;
4188 switch (nl_attr_type(a)) {
4189 case OVS_ACTION_ATTR_OUTPUT:
4190 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4191 if (port && port->bundle && port->bundle->bond) {
4192 bond_account(port->bundle->bond, &facet->flow,
4193 vlan_tci_to_vid(vlan_tci), n_bytes);
4197 case OVS_ACTION_ATTR_POP_VLAN:
4198 vlan_tci = htons(0);
4201 case OVS_ACTION_ATTR_PUSH_VLAN:
4202 vlan = nl_attr_get(a);
4203 vlan_tci = vlan->vlan_tci;
4209 /* Returns true if the only action for 'facet' is to send to the controller.
4210 * (We don't report NetFlow expiration messages for such facets because they
4211 * are just part of the control logic for the network, not real traffic). */
4213 facet_is_controller_flow(struct facet *facet)
4216 const struct rule *rule = &facet->rule->up;
4217 const struct ofpact *ofpacts = rule->ofpacts;
4218 size_t ofpacts_len = rule->ofpacts_len;
4220 if (ofpacts_len > 0 &&
4221 ofpacts->type == OFPACT_CONTROLLER &&
4222 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4229 /* Folds all of 'facet''s statistics into its rule. Also updates the
4230 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4231 * 'facet''s statistics in the datapath should have been zeroed and folded into
4232 * its packet and byte counts before this function is called. */
4234 facet_flush_stats(struct facet *facet)
4236 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4237 struct subfacet *subfacet;
4239 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4240 assert(!subfacet->dp_byte_count);
4241 assert(!subfacet->dp_packet_count);
4244 facet_push_stats(facet);
4245 if (facet->accounted_bytes < facet->byte_count) {
4246 facet_account(facet);
4247 facet->accounted_bytes = facet->byte_count;
4250 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4251 struct ofexpired expired;
4252 expired.flow = facet->flow;
4253 expired.packet_count = facet->packet_count;
4254 expired.byte_count = facet->byte_count;
4255 expired.used = facet->used;
4256 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4259 facet->rule->packet_count += facet->packet_count;
4260 facet->rule->byte_count += facet->byte_count;
4262 /* Reset counters to prevent double counting if 'facet' ever gets
4264 facet_reset_counters(facet);
4266 netflow_flow_clear(&facet->nf_flow);
4267 facet->tcp_flags = 0;
4270 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4271 * Returns it if found, otherwise a null pointer.
4273 * 'hash' must be the return value of flow_hash(flow, 0).
4275 * The returned facet might need revalidation; use facet_lookup_valid()
4276 * instead if that is important. */
4277 static struct facet *
4278 facet_find(struct ofproto_dpif *ofproto,
4279 const struct flow *flow, uint32_t hash)
4281 struct facet *facet;
4283 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4284 if (flow_equal(flow, &facet->flow)) {
4292 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4293 * Returns it if found, otherwise a null pointer.
4295 * 'hash' must be the return value of flow_hash(flow, 0).
4297 * The returned facet is guaranteed to be valid. */
4298 static struct facet *
4299 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4302 struct facet *facet;
4304 facet = facet_find(ofproto, flow, hash);
4306 && (ofproto->need_revalidate
4307 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
4308 facet_revalidate(facet);
4315 subfacet_path_to_string(enum subfacet_path path)
4318 case SF_NOT_INSTALLED:
4319 return "not installed";
4321 return "in fast path";
4323 return "in slow path";
4329 /* Returns the path in which a subfacet should be installed if its 'slow'
4330 * member has the specified value. */
4331 static enum subfacet_path
4332 subfacet_want_path(enum slow_path_reason slow)
4334 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4337 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4338 * supposing that its actions have been recalculated as 'want_actions' and that
4339 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4341 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4342 const struct ofpbuf *want_actions)
4344 enum subfacet_path want_path = subfacet_want_path(slow);
4345 return (want_path != subfacet->path
4346 || (want_path == SF_FAST_PATH
4347 && (subfacet->actions_len != want_actions->size
4348 || memcmp(subfacet->actions, want_actions->data,
4349 subfacet->actions_len))));
4353 facet_check_consistency(struct facet *facet)
4355 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4357 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4359 uint64_t odp_actions_stub[1024 / 8];
4360 struct ofpbuf odp_actions;
4362 struct rule_dpif *rule;
4363 struct subfacet *subfacet;
4364 bool may_log = false;
4367 /* Check the rule for consistency. */
4368 rule = rule_dpif_lookup(ofproto, &facet->flow);
4369 ok = rule == facet->rule;
4371 may_log = !VLOG_DROP_WARN(&rl);
4376 flow_format(&s, &facet->flow);
4377 ds_put_format(&s, ": facet associated with wrong rule (was "
4378 "table=%"PRIu8",", facet->rule->up.table_id);
4379 cls_rule_format(&facet->rule->up.cr, &s);
4380 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4382 cls_rule_format(&rule->up.cr, &s);
4383 ds_put_char(&s, ')');
4385 VLOG_WARN("%s", ds_cstr(&s));
4390 /* Check the datapath actions for consistency. */
4391 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4392 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4393 enum subfacet_path want_path;
4394 struct odputil_keybuf keybuf;
4395 struct action_xlate_ctx ctx;
4399 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4400 subfacet->initial_tci, rule, 0, NULL);
4401 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4404 if (subfacet->path == SF_NOT_INSTALLED) {
4405 /* This only happens if the datapath reported an error when we
4406 * tried to install the flow. Don't flag another error here. */
4410 want_path = subfacet_want_path(subfacet->slow);
4411 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4412 /* The actions for slow-path flows may legitimately vary from one
4413 * packet to the next. We're done. */
4417 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4421 /* Inconsistency! */
4423 may_log = !VLOG_DROP_WARN(&rl);
4427 /* Rate-limited, skip reporting. */
4432 subfacet_get_key(subfacet, &keybuf, &key);
4433 odp_flow_key_format(key.data, key.size, &s);
4435 ds_put_cstr(&s, ": inconsistency in subfacet");
4436 if (want_path != subfacet->path) {
4437 enum odp_key_fitness fitness = subfacet->key_fitness;
4439 ds_put_format(&s, " (%s, fitness=%s)",
4440 subfacet_path_to_string(subfacet->path),
4441 odp_key_fitness_to_string(fitness));
4442 ds_put_format(&s, " (should have been %s)",
4443 subfacet_path_to_string(want_path));
4444 } else if (want_path == SF_FAST_PATH) {
4445 ds_put_cstr(&s, " (actions were: ");
4446 format_odp_actions(&s, subfacet->actions,
4447 subfacet->actions_len);
4448 ds_put_cstr(&s, ") (correct actions: ");
4449 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4450 ds_put_char(&s, ')');
4452 ds_put_cstr(&s, " (actions: ");
4453 format_odp_actions(&s, subfacet->actions,
4454 subfacet->actions_len);
4455 ds_put_char(&s, ')');
4457 VLOG_WARN("%s", ds_cstr(&s));
4460 ofpbuf_uninit(&odp_actions);
4465 /* Re-searches the classifier for 'facet':
4467 * - If the rule found is different from 'facet''s current rule, moves
4468 * 'facet' to the new rule and recompiles its actions.
4470 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4471 * where it is and recompiles its actions anyway. */
4473 facet_revalidate(struct facet *facet)
4475 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4477 struct nlattr *odp_actions;
4480 struct actions *new_actions;
4482 struct action_xlate_ctx ctx;
4483 uint64_t odp_actions_stub[1024 / 8];
4484 struct ofpbuf odp_actions;
4486 struct rule_dpif *new_rule;
4487 struct subfacet *subfacet;
4490 COVERAGE_INC(facet_revalidate);
4492 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4494 /* Calculate new datapath actions.
4496 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4497 * emit a NetFlow expiration and, if so, we need to have the old state
4498 * around to properly compose it. */
4500 /* If the datapath actions changed or the installability changed,
4501 * then we need to talk to the datapath. */
4504 memset(&ctx, 0, sizeof ctx);
4505 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4506 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4507 enum slow_path_reason slow;
4509 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4510 subfacet->initial_tci, new_rule, 0, NULL);
4511 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4514 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4515 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4516 struct dpif_flow_stats stats;
4518 subfacet_install(subfacet,
4519 odp_actions.data, odp_actions.size, &stats, slow);
4520 subfacet_update_stats(subfacet, &stats);
4523 new_actions = xcalloc(list_size(&facet->subfacets),
4524 sizeof *new_actions);
4526 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4528 new_actions[i].actions_len = odp_actions.size;
4533 ofpbuf_uninit(&odp_actions);
4536 facet_flush_stats(facet);
4539 /* Update 'facet' now that we've taken care of all the old state. */
4540 facet->tags = ctx.tags;
4541 facet->nf_flow.output_iface = ctx.nf_output_iface;
4542 facet->has_learn = ctx.has_learn;
4543 facet->has_normal = ctx.has_normal;
4544 facet->has_fin_timeout = ctx.has_fin_timeout;
4545 facet->mirrors = ctx.mirrors;
4548 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4549 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4551 if (new_actions && new_actions[i].odp_actions) {
4552 free(subfacet->actions);
4553 subfacet->actions = new_actions[i].odp_actions;
4554 subfacet->actions_len = new_actions[i].actions_len;
4560 if (facet->rule != new_rule) {
4561 COVERAGE_INC(facet_changed_rule);
4562 list_remove(&facet->list_node);
4563 list_push_back(&new_rule->facets, &facet->list_node);
4564 facet->rule = new_rule;
4565 facet->used = new_rule->up.created;
4566 facet->prev_used = facet->used;
4570 /* Updates 'facet''s used time. Caller is responsible for calling
4571 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4573 facet_update_time(struct facet *facet, long long int used)
4575 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4576 if (used > facet->used) {
4578 ofproto_rule_update_used(&facet->rule->up, used);
4579 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4584 facet_reset_counters(struct facet *facet)
4586 facet->packet_count = 0;
4587 facet->byte_count = 0;
4588 facet->prev_packet_count = 0;
4589 facet->prev_byte_count = 0;
4590 facet->accounted_bytes = 0;
4594 facet_push_stats(struct facet *facet)
4596 struct dpif_flow_stats stats;
4598 assert(facet->packet_count >= facet->prev_packet_count);
4599 assert(facet->byte_count >= facet->prev_byte_count);
4600 assert(facet->used >= facet->prev_used);
4602 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4603 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4604 stats.used = facet->used;
4605 stats.tcp_flags = 0;
4607 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4608 facet->prev_packet_count = facet->packet_count;
4609 facet->prev_byte_count = facet->byte_count;
4610 facet->prev_used = facet->used;
4612 flow_push_stats(facet->rule, &facet->flow, &stats);
4614 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4615 facet->mirrors, stats.n_packets, stats.n_bytes);
4620 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4622 rule->packet_count += stats->n_packets;
4623 rule->byte_count += stats->n_bytes;
4624 ofproto_rule_update_used(&rule->up, stats->used);
4627 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4628 * 'rule''s actions and mirrors. */
4630 flow_push_stats(struct rule_dpif *rule,
4631 const struct flow *flow, const struct dpif_flow_stats *stats)
4633 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4634 struct action_xlate_ctx ctx;
4636 ofproto_rule_update_used(&rule->up, stats->used);
4638 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4640 ctx.resubmit_stats = stats;
4641 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4642 rule->up.ofpacts_len);
4647 static struct subfacet *
4648 subfacet_find(struct ofproto_dpif *ofproto,
4649 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4650 const struct flow *flow)
4652 struct subfacet *subfacet;
4654 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4655 &ofproto->subfacets) {
4657 ? (subfacet->key_len == key_len
4658 && !memcmp(key, subfacet->key, key_len))
4659 : flow_equal(flow, &subfacet->facet->flow)) {
4667 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4668 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4669 * existing subfacet if there is one, otherwise creates and returns a
4672 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4673 * which case the caller must populate the actions with
4674 * subfacet_make_actions(). */
4675 static struct subfacet *
4676 subfacet_create(struct facet *facet, struct flow_miss *miss,
4679 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4680 enum odp_key_fitness key_fitness = miss->key_fitness;
4681 const struct nlattr *key = miss->key;
4682 size_t key_len = miss->key_len;
4684 struct subfacet *subfacet;
4686 key_hash = odp_flow_key_hash(key, key_len);
4688 if (list_is_empty(&facet->subfacets)) {
4689 subfacet = &facet->one_subfacet;
4691 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4694 if (subfacet->facet == facet) {
4698 /* This shouldn't happen. */
4699 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4700 subfacet_destroy(subfacet);
4703 subfacet = xmalloc(sizeof *subfacet);
4706 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4707 list_push_back(&facet->subfacets, &subfacet->list_node);
4708 subfacet->facet = facet;
4709 subfacet->key_fitness = key_fitness;
4710 if (key_fitness != ODP_FIT_PERFECT) {
4711 subfacet->key = xmemdup(key, key_len);
4712 subfacet->key_len = key_len;
4714 subfacet->key = NULL;
4715 subfacet->key_len = 0;
4717 subfacet->used = now;
4718 subfacet->dp_packet_count = 0;
4719 subfacet->dp_byte_count = 0;
4720 subfacet->actions_len = 0;
4721 subfacet->actions = NULL;
4722 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4725 subfacet->path = SF_NOT_INSTALLED;
4726 subfacet->initial_tci = miss->initial_tci;
4727 subfacet->odp_in_port = miss->odp_in_port;
4732 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4733 * its facet within 'ofproto', and frees it. */
4735 subfacet_destroy__(struct subfacet *subfacet)
4737 struct facet *facet = subfacet->facet;
4738 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4740 subfacet_uninstall(subfacet);
4741 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4742 list_remove(&subfacet->list_node);
4743 free(subfacet->key);
4744 free(subfacet->actions);
4745 if (subfacet != &facet->one_subfacet) {
4750 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4751 * last remaining subfacet in its facet destroys the facet too. */
4753 subfacet_destroy(struct subfacet *subfacet)
4755 struct facet *facet = subfacet->facet;
4757 if (list_is_singleton(&facet->subfacets)) {
4758 /* facet_remove() needs at least one subfacet (it will remove it). */
4759 facet_remove(facet);
4761 subfacet_destroy__(subfacet);
4766 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
4767 struct subfacet **subfacets, int n)
4769 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
4770 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4771 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4772 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
4773 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4776 for (i = 0; i < n; i++) {
4777 ops[i].type = DPIF_OP_FLOW_DEL;
4778 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
4779 ops[i].u.flow_del.key = keys[i].data;
4780 ops[i].u.flow_del.key_len = keys[i].size;
4781 ops[i].u.flow_del.stats = &stats[i];
4785 dpif_operate(ofproto->backer->dpif, opsp, n);
4786 for (i = 0; i < n; i++) {
4787 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4788 subfacets[i]->path = SF_NOT_INSTALLED;
4789 subfacet_destroy(subfacets[i]);
4793 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4794 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4795 * for use as temporary storage. */
4797 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4801 if (!subfacet->key) {
4802 struct flow *flow = &subfacet->facet->flow;
4804 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4805 odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
4807 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4811 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4812 * Translates the actions into 'odp_actions', which the caller must have
4813 * initialized and is responsible for uninitializing. */
4815 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4816 struct ofpbuf *odp_actions)
4818 struct facet *facet = subfacet->facet;
4819 struct rule_dpif *rule = facet->rule;
4820 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4822 struct action_xlate_ctx ctx;
4824 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4826 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4827 facet->tags = ctx.tags;
4828 facet->has_learn = ctx.has_learn;
4829 facet->has_normal = ctx.has_normal;
4830 facet->has_fin_timeout = ctx.has_fin_timeout;
4831 facet->nf_flow.output_iface = ctx.nf_output_iface;
4832 facet->mirrors = ctx.mirrors;
4834 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4835 if (subfacet->actions_len != odp_actions->size
4836 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4837 free(subfacet->actions);
4838 subfacet->actions_len = odp_actions->size;
4839 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4843 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4844 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4845 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4846 * since 'subfacet' was last updated.
4848 * Returns 0 if successful, otherwise a positive errno value. */
4850 subfacet_install(struct subfacet *subfacet,
4851 const struct nlattr *actions, size_t actions_len,
4852 struct dpif_flow_stats *stats,
4853 enum slow_path_reason slow)
4855 struct facet *facet = subfacet->facet;
4856 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4857 enum subfacet_path path = subfacet_want_path(slow);
4858 uint64_t slow_path_stub[128 / 8];
4859 struct odputil_keybuf keybuf;
4860 enum dpif_flow_put_flags flags;
4864 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4866 flags |= DPIF_FP_ZERO_STATS;
4869 if (path == SF_SLOW_PATH) {
4870 compose_slow_path(ofproto, &facet->flow, slow,
4871 slow_path_stub, sizeof slow_path_stub,
4872 &actions, &actions_len);
4875 subfacet_get_key(subfacet, &keybuf, &key);
4876 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
4877 actions, actions_len, stats);
4880 subfacet_reset_dp_stats(subfacet, stats);
4884 subfacet->path = path;
4890 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4892 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4893 stats, subfacet->slow);
4896 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4898 subfacet_uninstall(struct subfacet *subfacet)
4900 if (subfacet->path != SF_NOT_INSTALLED) {
4901 struct rule_dpif *rule = subfacet->facet->rule;
4902 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4903 struct odputil_keybuf keybuf;
4904 struct dpif_flow_stats stats;
4908 subfacet_get_key(subfacet, &keybuf, &key);
4909 error = dpif_flow_del(ofproto->backer->dpif,
4910 key.data, key.size, &stats);
4911 subfacet_reset_dp_stats(subfacet, &stats);
4913 subfacet_update_stats(subfacet, &stats);
4915 subfacet->path = SF_NOT_INSTALLED;
4917 assert(subfacet->dp_packet_count == 0);
4918 assert(subfacet->dp_byte_count == 0);
4922 /* Resets 'subfacet''s datapath statistics counters. This should be called
4923 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4924 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4925 * was reset in the datapath. 'stats' will be modified to include only
4926 * statistics new since 'subfacet' was last updated. */
4928 subfacet_reset_dp_stats(struct subfacet *subfacet,
4929 struct dpif_flow_stats *stats)
4932 && subfacet->dp_packet_count <= stats->n_packets
4933 && subfacet->dp_byte_count <= stats->n_bytes) {
4934 stats->n_packets -= subfacet->dp_packet_count;
4935 stats->n_bytes -= subfacet->dp_byte_count;
4938 subfacet->dp_packet_count = 0;
4939 subfacet->dp_byte_count = 0;
4942 /* Updates 'subfacet''s used time. The caller is responsible for calling
4943 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4945 subfacet_update_time(struct subfacet *subfacet, long long int used)
4947 if (used > subfacet->used) {
4948 subfacet->used = used;
4949 facet_update_time(subfacet->facet, used);
4953 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4955 * Because of the meaning of a subfacet's counters, it only makes sense to do
4956 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4957 * represents a packet that was sent by hand or if it represents statistics
4958 * that have been cleared out of the datapath. */
4960 subfacet_update_stats(struct subfacet *subfacet,
4961 const struct dpif_flow_stats *stats)
4963 if (stats->n_packets || stats->used > subfacet->used) {
4964 struct facet *facet = subfacet->facet;
4966 subfacet_update_time(subfacet, stats->used);
4967 facet->packet_count += stats->n_packets;
4968 facet->byte_count += stats->n_bytes;
4969 facet->tcp_flags |= stats->tcp_flags;
4970 facet_push_stats(facet);
4971 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4977 static struct rule_dpif *
4978 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4980 struct rule_dpif *rule;
4982 rule = rule_dpif_lookup__(ofproto, flow, 0);
4987 return rule_dpif_miss_rule(ofproto, flow);
4990 static struct rule_dpif *
4991 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4994 struct cls_rule *cls_rule;
4995 struct classifier *cls;
4997 if (table_id >= N_TABLES) {
5001 cls = &ofproto->up.tables[table_id].cls;
5002 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5003 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5004 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5005 * are unavailable. */
5006 struct flow ofpc_normal_flow = *flow;
5007 ofpc_normal_flow.tp_src = htons(0);
5008 ofpc_normal_flow.tp_dst = htons(0);
5009 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5011 cls_rule = classifier_lookup(cls, flow);
5013 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5016 static struct rule_dpif *
5017 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5019 struct ofport_dpif *port;
5021 port = get_ofp_port(ofproto, flow->in_port);
5023 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5024 return ofproto->miss_rule;
5027 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5028 return ofproto->no_packet_in_rule;
5030 return ofproto->miss_rule;
5034 complete_operation(struct rule_dpif *rule)
5036 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5038 rule_invalidate(rule);
5040 struct dpif_completion *c = xmalloc(sizeof *c);
5041 c->op = rule->up.pending;
5042 list_push_back(&ofproto->completions, &c->list_node);
5044 ofoperation_complete(rule->up.pending, 0);
5048 static struct rule *
5051 struct rule_dpif *rule = xmalloc(sizeof *rule);
5056 rule_dealloc(struct rule *rule_)
5058 struct rule_dpif *rule = rule_dpif_cast(rule_);
5063 rule_construct(struct rule *rule_)
5065 struct rule_dpif *rule = rule_dpif_cast(rule_);
5066 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5067 struct rule_dpif *victim;
5070 rule->packet_count = 0;
5071 rule->byte_count = 0;
5073 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5074 if (victim && !list_is_empty(&victim->facets)) {
5075 struct facet *facet;
5077 rule->facets = victim->facets;
5078 list_moved(&rule->facets);
5079 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5080 /* XXX: We're only clearing our local counters here. It's possible
5081 * that quite a few packets are unaccounted for in the datapath
5082 * statistics. These will be accounted to the new rule instead of
5083 * cleared as required. This could be fixed by clearing out the
5084 * datapath statistics for this facet, but currently it doesn't
5086 facet_reset_counters(facet);
5090 /* Must avoid list_moved() in this case. */
5091 list_init(&rule->facets);
5094 table_id = rule->up.table_id;
5096 rule->tag = victim->tag;
5097 } else if (table_id == 0) {
5102 miniflow_expand(&rule->up.cr.match.flow, &flow);
5103 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5104 ofproto->tables[table_id].basis);
5107 complete_operation(rule);
5112 rule_destruct(struct rule *rule_)
5114 struct rule_dpif *rule = rule_dpif_cast(rule_);
5115 struct facet *facet, *next_facet;
5117 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5118 facet_revalidate(facet);
5121 complete_operation(rule);
5125 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5127 struct rule_dpif *rule = rule_dpif_cast(rule_);
5128 struct facet *facet;
5130 /* Start from historical data for 'rule' itself that are no longer tracked
5131 * in facets. This counts, for example, facets that have expired. */
5132 *packets = rule->packet_count;
5133 *bytes = rule->byte_count;
5135 /* Add any statistics that are tracked by facets. This includes
5136 * statistical data recently updated by ofproto_update_stats() as well as
5137 * stats for packets that were executed "by hand" via dpif_execute(). */
5138 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5139 *packets += facet->packet_count;
5140 *bytes += facet->byte_count;
5145 rule_execute(struct rule *rule_, const struct flow *flow,
5146 struct ofpbuf *packet)
5148 struct rule_dpif *rule = rule_dpif_cast(rule_);
5149 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5151 struct dpif_flow_stats stats;
5153 struct action_xlate_ctx ctx;
5154 uint64_t odp_actions_stub[1024 / 8];
5155 struct ofpbuf odp_actions;
5157 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5158 rule_credit_stats(rule, &stats);
5160 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5161 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5162 rule, stats.tcp_flags, packet);
5163 ctx.resubmit_stats = &stats;
5164 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5166 execute_odp_actions(ofproto, flow, odp_actions.data,
5167 odp_actions.size, packet);
5169 ofpbuf_uninit(&odp_actions);
5175 rule_modify_actions(struct rule *rule_)
5177 struct rule_dpif *rule = rule_dpif_cast(rule_);
5179 complete_operation(rule);
5182 /* Sends 'packet' out 'ofport'.
5183 * May modify 'packet'.
5184 * Returns 0 if successful, otherwise a positive errno value. */
5186 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5188 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5189 struct ofpbuf key, odp_actions;
5190 struct odputil_keybuf keybuf;
5195 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5196 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5198 if (odp_port != ofport->odp_port) {
5199 eth_pop_vlan(packet);
5200 flow.vlan_tci = htons(0);
5203 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5204 odp_flow_key_from_flow(&key, &flow,
5205 ofp_port_to_odp_port(ofproto, flow.in_port));
5207 ofpbuf_init(&odp_actions, 32);
5208 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5210 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5211 error = dpif_execute(ofproto->backer->dpif,
5213 odp_actions.data, odp_actions.size,
5215 ofpbuf_uninit(&odp_actions);
5218 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5219 ofproto->up.name, odp_port, strerror(error));
5221 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5225 /* OpenFlow to datapath action translation. */
5227 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5228 struct action_xlate_ctx *);
5229 static void xlate_normal(struct action_xlate_ctx *);
5231 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5232 * The action will state 'slow' as the reason that the action is in the slow
5233 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5234 * dump-flows" output to see why a flow is in the slow path.)
5236 * The 'stub_size' bytes in 'stub' will be used to store the action.
5237 * 'stub_size' must be large enough for the action.
5239 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5242 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5243 enum slow_path_reason slow,
5244 uint64_t *stub, size_t stub_size,
5245 const struct nlattr **actionsp, size_t *actions_lenp)
5247 union user_action_cookie cookie;
5250 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5251 cookie.slow_path.unused = 0;
5252 cookie.slow_path.reason = slow;
5254 ofpbuf_use_stack(&buf, stub, stub_size);
5255 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5256 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT16_MAX);
5257 odp_put_userspace_action(pid, &cookie, &buf);
5259 put_userspace_action(ofproto, &buf, flow, &cookie);
5261 *actionsp = buf.data;
5262 *actions_lenp = buf.size;
5266 put_userspace_action(const struct ofproto_dpif *ofproto,
5267 struct ofpbuf *odp_actions,
5268 const struct flow *flow,
5269 const union user_action_cookie *cookie)
5273 pid = dpif_port_get_pid(ofproto->backer->dpif,
5274 ofp_port_to_odp_port(ofproto, flow->in_port));
5276 return odp_put_userspace_action(pid, cookie, odp_actions);
5280 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5281 ovs_be16 vlan_tci, uint32_t odp_port,
5282 unsigned int n_outputs, union user_action_cookie *cookie)
5286 cookie->type = USER_ACTION_COOKIE_SFLOW;
5287 cookie->sflow.vlan_tci = vlan_tci;
5289 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5290 * port information") for the interpretation of cookie->output. */
5291 switch (n_outputs) {
5293 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5294 cookie->sflow.output = 0x40000000 | 256;
5298 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5300 cookie->sflow.output = ifindex;
5305 /* 0x80000000 means "multiple output ports. */
5306 cookie->sflow.output = 0x80000000 | n_outputs;
5311 /* Compose SAMPLE action for sFlow. */
5313 compose_sflow_action(const struct ofproto_dpif *ofproto,
5314 struct ofpbuf *odp_actions,
5315 const struct flow *flow,
5318 uint32_t probability;
5319 union user_action_cookie cookie;
5320 size_t sample_offset, actions_offset;
5323 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5327 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5329 /* Number of packets out of UINT_MAX to sample. */
5330 probability = dpif_sflow_get_probability(ofproto->sflow);
5331 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5333 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5334 compose_sflow_cookie(ofproto, htons(0), odp_port,
5335 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5336 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5338 nl_msg_end_nested(odp_actions, actions_offset);
5339 nl_msg_end_nested(odp_actions, sample_offset);
5340 return cookie_offset;
5343 /* SAMPLE action must be first action in any given list of actions.
5344 * At this point we do not have all information required to build it. So try to
5345 * build sample action as complete as possible. */
5347 add_sflow_action(struct action_xlate_ctx *ctx)
5349 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5351 &ctx->flow, OVSP_NONE);
5352 ctx->sflow_odp_port = 0;
5353 ctx->sflow_n_outputs = 0;
5356 /* Fix SAMPLE action according to data collected while composing ODP actions.
5357 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5358 * USERSPACE action's user-cookie which is required for sflow. */
5360 fix_sflow_action(struct action_xlate_ctx *ctx)
5362 const struct flow *base = &ctx->base_flow;
5363 union user_action_cookie *cookie;
5365 if (!ctx->user_cookie_offset) {
5369 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5371 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5373 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5374 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5378 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5381 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5382 uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5383 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5384 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5388 struct priority_to_dscp *pdscp;
5390 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5391 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5393 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5394 xlate_report(ctx, "STP not in forwarding state, skipping output");
5398 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5400 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5401 ctx->flow.nw_tos |= pdscp->dscp;
5404 /* We may not have an ofport record for this port, but it doesn't hurt
5405 * to allow forwarding to it anyhow. Maybe such a port will appear
5406 * later and we're pre-populating the flow table. */
5409 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5410 ctx->flow.vlan_tci);
5411 if (out_port != odp_port) {
5412 ctx->flow.vlan_tci = htons(0);
5414 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5415 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5417 ctx->sflow_odp_port = odp_port;
5418 ctx->sflow_n_outputs++;
5419 ctx->nf_output_iface = ofp_port;
5420 ctx->flow.vlan_tci = flow_vlan_tci;
5421 ctx->flow.nw_tos = flow_nw_tos;
5425 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5427 compose_output_action__(ctx, ofp_port, true);
5431 xlate_table_action(struct action_xlate_ctx *ctx,
5432 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5434 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5435 struct ofproto_dpif *ofproto = ctx->ofproto;
5436 struct rule_dpif *rule;
5437 uint16_t old_in_port;
5438 uint8_t old_table_id;
5440 old_table_id = ctx->table_id;
5441 ctx->table_id = table_id;
5443 /* Look up a flow with 'in_port' as the input port. */
5444 old_in_port = ctx->flow.in_port;
5445 ctx->flow.in_port = in_port;
5446 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5449 if (table_id > 0 && table_id < N_TABLES) {
5450 struct table_dpif *table = &ofproto->tables[table_id];
5451 if (table->other_table) {
5452 ctx->tags |= (rule && rule->tag
5454 : rule_calculate_tag(&ctx->flow,
5455 &table->other_table->mask,
5460 /* Restore the original input port. Otherwise OFPP_NORMAL and
5461 * OFPP_IN_PORT will have surprising behavior. */
5462 ctx->flow.in_port = old_in_port;
5464 if (ctx->resubmit_hook) {
5465 ctx->resubmit_hook(ctx, rule);
5468 if (rule == NULL && may_packet_in) {
5470 * check if table configuration flags
5471 * OFPTC_TABLE_MISS_CONTROLLER, default.
5472 * OFPTC_TABLE_MISS_CONTINUE,
5473 * OFPTC_TABLE_MISS_DROP
5474 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5476 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5480 struct rule_dpif *old_rule = ctx->rule;
5482 if (ctx->resubmit_stats) {
5483 rule_credit_stats(rule, ctx->resubmit_stats);
5488 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5489 ctx->rule = old_rule;
5493 ctx->table_id = old_table_id;
5495 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5497 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5498 MAX_RESUBMIT_RECURSION);
5499 ctx->max_resubmit_trigger = true;
5504 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5505 const struct ofpact_resubmit *resubmit)
5510 in_port = resubmit->in_port;
5511 if (in_port == OFPP_IN_PORT) {
5512 in_port = ctx->flow.in_port;
5515 table_id = resubmit->table_id;
5516 if (table_id == 255) {
5517 table_id = ctx->table_id;
5520 xlate_table_action(ctx, in_port, table_id, false);
5524 flood_packets(struct action_xlate_ctx *ctx, bool all)
5526 struct ofport_dpif *ofport;
5528 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5529 uint16_t ofp_port = ofport->up.ofp_port;
5531 if (ofp_port == ctx->flow.in_port) {
5536 compose_output_action__(ctx, ofp_port, false);
5537 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5538 compose_output_action(ctx, ofp_port);
5542 ctx->nf_output_iface = NF_OUT_FLOOD;
5546 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5547 enum ofp_packet_in_reason reason,
5548 uint16_t controller_id)
5550 struct ofputil_packet_in pin;
5551 struct ofpbuf *packet;
5553 ctx->slow |= SLOW_CONTROLLER;
5558 packet = ofpbuf_clone(ctx->packet);
5560 if (packet->l2 && packet->l3) {
5561 struct eth_header *eh;
5563 eth_pop_vlan(packet);
5566 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5567 * LLC frame. Calculating the Ethernet type of these frames is more
5568 * trouble than seems appropriate for a simple assertion. */
5569 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5570 || eh->eth_type == ctx->flow.dl_type);
5572 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5573 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5575 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5576 eth_push_vlan(packet, ctx->flow.vlan_tci);
5580 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5581 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5582 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5586 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5587 packet_set_tcp_port(packet, ctx->flow.tp_src,
5589 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5590 packet_set_udp_port(packet, ctx->flow.tp_src,
5597 pin.packet = packet->data;
5598 pin.packet_len = packet->size;
5599 pin.reason = reason;
5600 pin.controller_id = controller_id;
5601 pin.table_id = ctx->table_id;
5602 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5605 flow_get_metadata(&ctx->flow, &pin.fmd);
5607 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5608 ofpbuf_delete(packet);
5612 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5614 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5615 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5619 if (ctx->flow.nw_ttl > 1) {
5625 for (i = 0; i < ids->n_controllers; i++) {
5626 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5630 /* Stop processing for current table. */
5636 xlate_output_action(struct action_xlate_ctx *ctx,
5637 uint16_t port, uint16_t max_len, bool may_packet_in)
5639 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5641 ctx->nf_output_iface = NF_OUT_DROP;
5645 compose_output_action(ctx, ctx->flow.in_port);
5648 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5654 flood_packets(ctx, false);
5657 flood_packets(ctx, true);
5659 case OFPP_CONTROLLER:
5660 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5666 if (port != ctx->flow.in_port) {
5667 compose_output_action(ctx, port);
5669 xlate_report(ctx, "skipping output to input port");
5674 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5675 ctx->nf_output_iface = NF_OUT_FLOOD;
5676 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5677 ctx->nf_output_iface = prev_nf_output_iface;
5678 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5679 ctx->nf_output_iface != NF_OUT_FLOOD) {
5680 ctx->nf_output_iface = NF_OUT_MULTI;
5685 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5686 const struct ofpact_output_reg *or)
5688 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5689 if (port <= UINT16_MAX) {
5690 xlate_output_action(ctx, port, or->max_len, false);
5695 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5696 const struct ofpact_enqueue *enqueue)
5698 uint16_t ofp_port = enqueue->port;
5699 uint32_t queue_id = enqueue->queue;
5700 uint32_t flow_priority, priority;
5703 /* Translate queue to priority. */
5704 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5705 queue_id, &priority);
5707 /* Fall back to ordinary output action. */
5708 xlate_output_action(ctx, enqueue->port, 0, false);
5712 /* Check output port. */
5713 if (ofp_port == OFPP_IN_PORT) {
5714 ofp_port = ctx->flow.in_port;
5715 } else if (ofp_port == ctx->flow.in_port) {
5719 /* Add datapath actions. */
5720 flow_priority = ctx->flow.skb_priority;
5721 ctx->flow.skb_priority = priority;
5722 compose_output_action(ctx, ofp_port);
5723 ctx->flow.skb_priority = flow_priority;
5725 /* Update NetFlow output port. */
5726 if (ctx->nf_output_iface == NF_OUT_DROP) {
5727 ctx->nf_output_iface = ofp_port;
5728 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5729 ctx->nf_output_iface = NF_OUT_MULTI;
5734 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5736 uint32_t skb_priority;
5738 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5739 queue_id, &skb_priority)) {
5740 ctx->flow.skb_priority = skb_priority;
5742 /* Couldn't translate queue to a priority. Nothing to do. A warning
5743 * has already been logged. */
5747 struct xlate_reg_state {
5753 xlate_autopath(struct action_xlate_ctx *ctx,
5754 const struct ofpact_autopath *ap)
5756 uint16_t ofp_port = ap->port;
5757 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5759 if (!port || !port->bundle) {
5760 ofp_port = OFPP_NONE;
5761 } else if (port->bundle->bond) {
5762 /* Autopath does not support VLAN hashing. */
5763 struct ofport_dpif *slave = bond_choose_output_slave(
5764 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5766 ofp_port = slave->up.ofp_port;
5769 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5773 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5775 struct ofproto_dpif *ofproto = ofproto_;
5776 struct ofport_dpif *port;
5786 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5789 port = get_ofp_port(ofproto, ofp_port);
5790 return port ? port->may_enable : false;
5795 xlate_bundle_action(struct action_xlate_ctx *ctx,
5796 const struct ofpact_bundle *bundle)
5800 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5801 if (bundle->dst.field) {
5802 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5804 xlate_output_action(ctx, port, 0, false);
5809 xlate_learn_action(struct action_xlate_ctx *ctx,
5810 const struct ofpact_learn *learn)
5812 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5813 struct ofputil_flow_mod fm;
5814 uint64_t ofpacts_stub[1024 / 8];
5815 struct ofpbuf ofpacts;
5818 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5819 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5821 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5822 if (error && !VLOG_DROP_WARN(&rl)) {
5823 VLOG_WARN("learning action failed to modify flow table (%s)",
5824 ofperr_get_name(error));
5827 ofpbuf_uninit(&ofpacts);
5830 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5831 * means "infinite". */
5833 reduce_timeout(uint16_t max, uint16_t *timeout)
5835 if (max && (!*timeout || *timeout > max)) {
5841 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5842 const struct ofpact_fin_timeout *oft)
5844 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5845 struct rule_dpif *rule = ctx->rule;
5847 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5848 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5853 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5855 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5856 ? OFPUTIL_PC_NO_RECV_STP
5857 : OFPUTIL_PC_NO_RECV)) {
5861 /* Only drop packets here if both forwarding and learning are
5862 * disabled. If just learning is enabled, we need to have
5863 * OFPP_NORMAL and the learning action have a look at the packet
5864 * before we can drop it. */
5865 if (!stp_forward_in_state(port->stp_state)
5866 && !stp_learn_in_state(port->stp_state)) {
5874 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5875 struct action_xlate_ctx *ctx)
5877 const struct ofport_dpif *port;
5878 bool was_evictable = true;
5879 const struct ofpact *a;
5881 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5882 if (port && !may_receive(port, ctx)) {
5883 /* Drop this flow. */
5888 /* Don't let the rule we're working on get evicted underneath us. */
5889 was_evictable = ctx->rule->up.evictable;
5890 ctx->rule->up.evictable = false;
5892 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5893 struct ofpact_controller *controller;
5894 const struct ofpact_metadata *metadata;
5902 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5903 ofpact_get_OUTPUT(a)->max_len, true);
5906 case OFPACT_CONTROLLER:
5907 controller = ofpact_get_CONTROLLER(a);
5908 execute_controller_action(ctx, controller->max_len,
5910 controller->controller_id);
5913 case OFPACT_ENQUEUE:
5914 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5917 case OFPACT_SET_VLAN_VID:
5918 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5919 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5923 case OFPACT_SET_VLAN_PCP:
5924 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5925 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5930 case OFPACT_STRIP_VLAN:
5931 ctx->flow.vlan_tci = htons(0);
5934 case OFPACT_PUSH_VLAN:
5935 /* TODO:XXX 802.1AD(QinQ) */
5936 ctx->flow.vlan_tci = htons(VLAN_CFI);
5939 case OFPACT_SET_ETH_SRC:
5940 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5944 case OFPACT_SET_ETH_DST:
5945 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5949 case OFPACT_SET_IPV4_SRC:
5950 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5953 case OFPACT_SET_IPV4_DST:
5954 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5957 case OFPACT_SET_IPV4_DSCP:
5958 /* OpenFlow 1.0 only supports IPv4. */
5959 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5960 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5961 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5965 case OFPACT_SET_L4_SRC_PORT:
5966 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5969 case OFPACT_SET_L4_DST_PORT:
5970 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5973 case OFPACT_RESUBMIT:
5974 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5977 case OFPACT_SET_TUNNEL:
5978 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5981 case OFPACT_SET_QUEUE:
5982 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5985 case OFPACT_POP_QUEUE:
5986 ctx->flow.skb_priority = ctx->orig_skb_priority;
5989 case OFPACT_REG_MOVE:
5990 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
5993 case OFPACT_REG_LOAD:
5994 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
5997 case OFPACT_DEC_TTL:
5998 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6004 /* Nothing to do. */
6007 case OFPACT_MULTIPATH:
6008 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6011 case OFPACT_AUTOPATH:
6012 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6016 ctx->ofproto->has_bundle_action = true;
6017 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6020 case OFPACT_OUTPUT_REG:
6021 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6025 ctx->has_learn = true;
6026 if (ctx->may_learn) {
6027 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6035 case OFPACT_FIN_TIMEOUT:
6036 ctx->has_fin_timeout = true;
6037 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6040 case OFPACT_CLEAR_ACTIONS:
6042 * Nothing to do because writa-actions is not supported for now.
6043 * When writa-actions is supported, clear-actions also must
6044 * be supported at the same time.
6048 case OFPACT_WRITE_METADATA:
6049 metadata = ofpact_get_WRITE_METADATA(a);
6050 ctx->flow.metadata &= ~metadata->mask;
6051 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6054 case OFPACT_GOTO_TABLE: {
6055 /* TODO:XXX remove recursion */
6056 /* It is assumed that goto-table is last action */
6057 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6058 assert(ctx->table_id < ogt->table_id);
6059 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6066 /* We've let OFPP_NORMAL and the learning action look at the packet,
6067 * so drop it now if forwarding is disabled. */
6068 if (port && !stp_forward_in_state(port->stp_state)) {
6069 ofpbuf_clear(ctx->odp_actions);
6070 add_sflow_action(ctx);
6073 ctx->rule->up.evictable = was_evictable;
6078 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6079 struct ofproto_dpif *ofproto, const struct flow *flow,
6080 ovs_be16 initial_tci, struct rule_dpif *rule,
6081 uint8_t tcp_flags, const struct ofpbuf *packet)
6083 ctx->ofproto = ofproto;
6085 ctx->base_flow = ctx->flow;
6086 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
6087 ctx->base_flow.vlan_tci = initial_tci;
6089 ctx->packet = packet;
6090 ctx->may_learn = packet != NULL;
6091 ctx->tcp_flags = tcp_flags;
6092 ctx->resubmit_hook = NULL;
6093 ctx->report_hook = NULL;
6094 ctx->resubmit_stats = NULL;
6097 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6098 * into datapath actions in 'odp_actions', using 'ctx'. */
6100 xlate_actions(struct action_xlate_ctx *ctx,
6101 const struct ofpact *ofpacts, size_t ofpacts_len,
6102 struct ofpbuf *odp_actions)
6104 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6105 * that in the future we always keep a copy of the original flow for
6106 * tracing purposes. */
6107 static bool hit_resubmit_limit;
6109 enum slow_path_reason special;
6111 COVERAGE_INC(ofproto_dpif_xlate);
6113 ofpbuf_clear(odp_actions);
6114 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6116 ctx->odp_actions = odp_actions;
6119 ctx->has_learn = false;
6120 ctx->has_normal = false;
6121 ctx->has_fin_timeout = false;
6122 ctx->nf_output_iface = NF_OUT_DROP;
6125 ctx->max_resubmit_trigger = false;
6126 ctx->orig_skb_priority = ctx->flow.skb_priority;
6130 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6131 /* Do this conditionally because the copy is expensive enough that it
6132 * shows up in profiles.
6134 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6135 * believe that I wasn't using it without initializing it if I kept it
6136 * in a local variable. */
6137 ctx->orig_flow = ctx->flow;
6140 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6141 switch (ctx->ofproto->up.frag_handling) {
6142 case OFPC_FRAG_NORMAL:
6143 /* We must pretend that transport ports are unavailable. */
6144 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6145 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6148 case OFPC_FRAG_DROP:
6151 case OFPC_FRAG_REASM:
6154 case OFPC_FRAG_NX_MATCH:
6155 /* Nothing to do. */
6158 case OFPC_INVALID_TTL_TO_CONTROLLER:
6163 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
6165 ctx->slow |= special;
6167 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6168 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6170 add_sflow_action(ctx);
6171 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6173 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6174 if (!hit_resubmit_limit) {
6175 /* We didn't record the original flow. Make sure we do from
6177 hit_resubmit_limit = true;
6178 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6179 struct ds ds = DS_EMPTY_INITIALIZER;
6181 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6183 VLOG_ERR("Trace triggered by excessive resubmit "
6184 "recursion:\n%s", ds_cstr(&ds));
6189 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6190 ctx->odp_actions->data,
6191 ctx->odp_actions->size)) {
6192 ctx->slow |= SLOW_IN_BAND;
6194 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6196 compose_output_action(ctx, OFPP_LOCAL);
6199 if (ctx->ofproto->has_mirrors) {
6200 add_mirror_actions(ctx, &ctx->orig_flow);
6202 fix_sflow_action(ctx);
6206 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6207 * into datapath actions, using 'ctx', and discards the datapath actions. */
6209 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6210 const struct ofpact *ofpacts,
6213 uint64_t odp_actions_stub[1024 / 8];
6214 struct ofpbuf odp_actions;
6216 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6217 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6218 ofpbuf_uninit(&odp_actions);
6222 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6224 if (ctx->report_hook) {
6225 ctx->report_hook(ctx, s);
6229 /* OFPP_NORMAL implementation. */
6231 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6233 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6234 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6235 * the bundle on which the packet was received, returns the VLAN to which the
6238 * Both 'vid' and the return value are in the range 0...4095. */
6240 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6242 switch (in_bundle->vlan_mode) {
6243 case PORT_VLAN_ACCESS:
6244 return in_bundle->vlan;
6247 case PORT_VLAN_TRUNK:
6250 case PORT_VLAN_NATIVE_UNTAGGED:
6251 case PORT_VLAN_NATIVE_TAGGED:
6252 return vid ? vid : in_bundle->vlan;
6259 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6260 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6263 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6264 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6267 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6269 /* Allow any VID on the OFPP_NONE port. */
6270 if (in_bundle == &ofpp_none_bundle) {
6274 switch (in_bundle->vlan_mode) {
6275 case PORT_VLAN_ACCESS:
6278 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6279 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6280 "packet received on port %s configured as VLAN "
6281 "%"PRIu16" access port",
6282 in_bundle->ofproto->up.name, vid,
6283 in_bundle->name, in_bundle->vlan);
6289 case PORT_VLAN_NATIVE_UNTAGGED:
6290 case PORT_VLAN_NATIVE_TAGGED:
6292 /* Port must always carry its native VLAN. */
6296 case PORT_VLAN_TRUNK:
6297 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6299 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6300 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6301 "received on port %s not configured for trunking "
6303 in_bundle->ofproto->up.name, vid,
6304 in_bundle->name, vid);
6316 /* Given 'vlan', the VLAN that a packet belongs to, and
6317 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6318 * that should be included in the 802.1Q header. (If the return value is 0,
6319 * then the 802.1Q header should only be included in the packet if there is a
6322 * Both 'vlan' and the return value are in the range 0...4095. */
6324 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6326 switch (out_bundle->vlan_mode) {
6327 case PORT_VLAN_ACCESS:
6330 case PORT_VLAN_TRUNK:
6331 case PORT_VLAN_NATIVE_TAGGED:
6334 case PORT_VLAN_NATIVE_UNTAGGED:
6335 return vlan == out_bundle->vlan ? 0 : vlan;
6343 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6346 struct ofport_dpif *port;
6348 ovs_be16 tci, old_tci;
6350 vid = output_vlan_to_vid(out_bundle, vlan);
6351 if (!out_bundle->bond) {
6352 port = ofbundle_get_a_port(out_bundle);
6354 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6357 /* No slaves enabled, so drop packet. */
6362 old_tci = ctx->flow.vlan_tci;
6364 if (tci || out_bundle->use_priority_tags) {
6365 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6367 tci |= htons(VLAN_CFI);
6370 ctx->flow.vlan_tci = tci;
6372 compose_output_action(ctx, port->up.ofp_port);
6373 ctx->flow.vlan_tci = old_tci;
6377 mirror_mask_ffs(mirror_mask_t mask)
6379 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6384 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6386 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6387 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6391 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6393 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6396 /* Returns an arbitrary interface within 'bundle'. */
6397 static struct ofport_dpif *
6398 ofbundle_get_a_port(const struct ofbundle *bundle)
6400 return CONTAINER_OF(list_front(&bundle->ports),
6401 struct ofport_dpif, bundle_node);
6405 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6407 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6411 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6413 struct ofproto_dpif *ofproto = ctx->ofproto;
6414 mirror_mask_t mirrors;
6415 struct ofbundle *in_bundle;
6418 const struct nlattr *a;
6421 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6422 ctx->packet != NULL, NULL);
6426 mirrors = in_bundle->src_mirrors;
6428 /* Drop frames on bundles reserved for mirroring. */
6429 if (in_bundle->mirror_out) {
6430 if (ctx->packet != NULL) {
6431 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6432 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6433 "%s, which is reserved exclusively for mirroring",
6434 ctx->ofproto->up.name, in_bundle->name);
6440 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6441 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6444 vlan = input_vid_to_vlan(in_bundle, vid);
6446 /* Look at the output ports to check for destination selections. */
6448 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6449 ctx->odp_actions->size) {
6450 enum ovs_action_attr type = nl_attr_type(a);
6451 struct ofport_dpif *ofport;
6453 if (type != OVS_ACTION_ATTR_OUTPUT) {
6457 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6458 if (ofport && ofport->bundle) {
6459 mirrors |= ofport->bundle->dst_mirrors;
6467 /* Restore the original packet before adding the mirror actions. */
6468 ctx->flow = *orig_flow;
6473 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6475 if (!vlan_is_mirrored(m, vlan)) {
6476 mirrors = zero_rightmost_1bit(mirrors);
6480 mirrors &= ~m->dup_mirrors;
6481 ctx->mirrors |= m->dup_mirrors;
6483 output_normal(ctx, m->out, vlan);
6484 } else if (vlan != m->out_vlan
6485 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6486 struct ofbundle *bundle;
6488 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6489 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6490 && !bundle->mirror_out) {
6491 output_normal(ctx, bundle, m->out_vlan);
6499 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6500 uint64_t packets, uint64_t bytes)
6506 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6509 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6512 /* In normal circumstances 'm' will not be NULL. However,
6513 * if mirrors are reconfigured, we can temporarily get out
6514 * of sync in facet_revalidate(). We could "correct" the
6515 * mirror list before reaching here, but doing that would
6516 * not properly account the traffic stats we've currently
6517 * accumulated for previous mirror configuration. */
6521 m->packet_count += packets;
6522 m->byte_count += bytes;
6526 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6527 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6528 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6530 is_gratuitous_arp(const struct flow *flow)
6532 return (flow->dl_type == htons(ETH_TYPE_ARP)
6533 && eth_addr_is_broadcast(flow->dl_dst)
6534 && (flow->nw_proto == ARP_OP_REPLY
6535 || (flow->nw_proto == ARP_OP_REQUEST
6536 && flow->nw_src == flow->nw_dst)));
6540 update_learning_table(struct ofproto_dpif *ofproto,
6541 const struct flow *flow, int vlan,
6542 struct ofbundle *in_bundle)
6544 struct mac_entry *mac;
6546 /* Don't learn the OFPP_NONE port. */
6547 if (in_bundle == &ofpp_none_bundle) {
6551 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6555 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6556 if (is_gratuitous_arp(flow)) {
6557 /* We don't want to learn from gratuitous ARP packets that are
6558 * reflected back over bond slaves so we lock the learning table. */
6559 if (!in_bundle->bond) {
6560 mac_entry_set_grat_arp_lock(mac);
6561 } else if (mac_entry_is_grat_arp_locked(mac)) {
6566 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6567 /* The log messages here could actually be useful in debugging,
6568 * so keep the rate limit relatively high. */
6569 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6570 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6571 "on port %s in VLAN %d",
6572 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6573 in_bundle->name, vlan);
6575 mac->port.p = in_bundle;
6576 tag_set_add(&ofproto->revalidate_set,
6577 mac_learning_changed(ofproto->ml, mac));
6581 static struct ofbundle *
6582 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6583 bool warn, struct ofport_dpif **in_ofportp)
6585 struct ofport_dpif *ofport;
6587 /* Find the port and bundle for the received packet. */
6588 ofport = get_ofp_port(ofproto, in_port);
6590 *in_ofportp = ofport;
6592 if (ofport && ofport->bundle) {
6593 return ofport->bundle;
6596 /* Special-case OFPP_NONE, which a controller may use as the ingress
6597 * port for traffic that it is sourcing. */
6598 if (in_port == OFPP_NONE) {
6599 return &ofpp_none_bundle;
6602 /* Odd. A few possible reasons here:
6604 * - We deleted a port but there are still a few packets queued up
6607 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6608 * we don't know about.
6610 * - The ofproto client didn't configure the port as part of a bundle.
6611 * This is particularly likely to happen if a packet was received on the
6612 * port after it was created, but before the client had a chance to
6613 * configure its bundle.
6616 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6618 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6619 "port %"PRIu16, ofproto->up.name, in_port);
6624 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6625 * dropped. Returns true if they may be forwarded, false if they should be
6628 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6629 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6631 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6632 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6633 * checked by input_vid_is_valid().
6635 * May also add tags to '*tags', although the current implementation only does
6636 * so in one special case.
6639 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6642 struct ofproto_dpif *ofproto = ctx->ofproto;
6643 struct flow *flow = &ctx->flow;
6644 struct ofbundle *in_bundle = in_port->bundle;
6646 /* Drop frames for reserved multicast addresses
6647 * only if forward_bpdu option is absent. */
6648 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6649 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6653 if (in_bundle->bond) {
6654 struct mac_entry *mac;
6656 switch (bond_check_admissibility(in_bundle->bond, in_port,
6657 flow->dl_dst, &ctx->tags)) {
6662 xlate_report(ctx, "bonding refused admissibility, dropping");
6665 case BV_DROP_IF_MOVED:
6666 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6667 if (mac && mac->port.p != in_bundle &&
6668 (!is_gratuitous_arp(flow)
6669 || mac_entry_is_grat_arp_locked(mac))) {
6670 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6682 xlate_normal(struct action_xlate_ctx *ctx)
6684 struct ofport_dpif *in_port;
6685 struct ofbundle *in_bundle;
6686 struct mac_entry *mac;
6690 ctx->has_normal = true;
6692 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6693 ctx->packet != NULL, &in_port);
6695 xlate_report(ctx, "no input bundle, dropping");
6699 /* Drop malformed frames. */
6700 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6701 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6702 if (ctx->packet != NULL) {
6703 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6704 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6705 "VLAN tag received on port %s",
6706 ctx->ofproto->up.name, in_bundle->name);
6708 xlate_report(ctx, "partial VLAN tag, dropping");
6712 /* Drop frames on bundles reserved for mirroring. */
6713 if (in_bundle->mirror_out) {
6714 if (ctx->packet != NULL) {
6715 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6716 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6717 "%s, which is reserved exclusively for mirroring",
6718 ctx->ofproto->up.name, in_bundle->name);
6720 xlate_report(ctx, "input port is mirror output port, dropping");
6725 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6726 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6727 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6730 vlan = input_vid_to_vlan(in_bundle, vid);
6732 /* Check other admissibility requirements. */
6733 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6737 /* Learn source MAC. */
6738 if (ctx->may_learn) {
6739 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6742 /* Determine output bundle. */
6743 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6746 if (mac->port.p != in_bundle) {
6747 xlate_report(ctx, "forwarding to learned port");
6748 output_normal(ctx, mac->port.p, vlan);
6750 xlate_report(ctx, "learned port is input port, dropping");
6753 struct ofbundle *bundle;
6755 xlate_report(ctx, "no learned MAC for destination, flooding");
6756 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6757 if (bundle != in_bundle
6758 && ofbundle_includes_vlan(bundle, vlan)
6759 && bundle->floodable
6760 && !bundle->mirror_out) {
6761 output_normal(ctx, bundle, vlan);
6764 ctx->nf_output_iface = NF_OUT_FLOOD;
6768 /* Optimized flow revalidation.
6770 * It's a difficult problem, in general, to tell which facets need to have
6771 * their actions recalculated whenever the OpenFlow flow table changes. We
6772 * don't try to solve that general problem: for most kinds of OpenFlow flow
6773 * table changes, we recalculate the actions for every facet. This is
6774 * relatively expensive, but it's good enough if the OpenFlow flow table
6775 * doesn't change very often.
6777 * However, we can expect one particular kind of OpenFlow flow table change to
6778 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6779 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6780 * table, we add a special case that applies to flow tables in which every rule
6781 * has the same form (that is, the same wildcards), except that the table is
6782 * also allowed to have a single "catch-all" flow that matches all packets. We
6783 * optimize this case by tagging all of the facets that resubmit into the table
6784 * and invalidating the same tag whenever a flow changes in that table. The
6785 * end result is that we revalidate just the facets that need it (and sometimes
6786 * a few more, but not all of the facets or even all of the facets that
6787 * resubmit to the table modified by MAC learning). */
6789 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6790 * into an OpenFlow table with the given 'basis'. */
6792 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
6795 if (minimask_is_catchall(mask)) {
6798 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
6799 return tag_create_deterministic(hash);
6803 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6804 * taggability of that table.
6806 * This function must be called after *each* change to a flow table. If you
6807 * skip calling it on some changes then the pointer comparisons at the end can
6808 * be invalid if you get unlucky. For example, if a flow removal causes a
6809 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6810 * different wildcards to be created with the same address, then this function
6811 * will incorrectly skip revalidation. */
6813 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6815 struct table_dpif *table = &ofproto->tables[table_id];
6816 const struct oftable *oftable = &ofproto->up.tables[table_id];
6817 struct cls_table *catchall, *other;
6818 struct cls_table *t;
6820 catchall = other = NULL;
6822 switch (hmap_count(&oftable->cls.tables)) {
6824 /* We could tag this OpenFlow table but it would make the logic a
6825 * little harder and it's a corner case that doesn't seem worth it
6831 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6832 if (cls_table_is_catchall(t)) {
6834 } else if (!other) {
6837 /* Indicate that we can't tag this by setting both tables to
6838 * NULL. (We know that 'catchall' is already NULL.) */
6845 /* Can't tag this table. */
6849 if (table->catchall_table != catchall || table->other_table != other) {
6850 table->catchall_table = catchall;
6851 table->other_table = other;
6852 ofproto->need_revalidate = REV_FLOW_TABLE;
6856 /* Given 'rule' that has changed in some way (either it is a rule being
6857 * inserted, a rule being deleted, or a rule whose actions are being
6858 * modified), marks facets for revalidation to ensure that packets will be
6859 * forwarded correctly according to the new state of the flow table.
6861 * This function must be called after *each* change to a flow table. See
6862 * the comment on table_update_taggable() for more information. */
6864 rule_invalidate(const struct rule_dpif *rule)
6866 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6868 table_update_taggable(ofproto, rule->up.table_id);
6870 if (!ofproto->need_revalidate) {
6871 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6873 if (table->other_table && rule->tag) {
6874 tag_set_add(&ofproto->revalidate_set, rule->tag);
6876 ofproto->need_revalidate = REV_FLOW_TABLE;
6882 set_frag_handling(struct ofproto *ofproto_,
6883 enum ofp_config_flags frag_handling)
6885 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6887 if (frag_handling != OFPC_FRAG_REASM) {
6888 ofproto->need_revalidate = REV_RECONFIGURE;
6896 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6897 const struct flow *flow,
6898 const struct ofpact *ofpacts, size_t ofpacts_len)
6900 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6901 struct odputil_keybuf keybuf;
6902 struct dpif_flow_stats stats;
6906 struct action_xlate_ctx ctx;
6907 uint64_t odp_actions_stub[1024 / 8];
6908 struct ofpbuf odp_actions;
6910 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6911 odp_flow_key_from_flow(&key, flow,
6912 ofp_port_to_odp_port(ofproto, flow->in_port));
6914 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6916 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6917 packet_get_tcp_flags(packet, flow), packet);
6918 ctx.resubmit_stats = &stats;
6920 ofpbuf_use_stub(&odp_actions,
6921 odp_actions_stub, sizeof odp_actions_stub);
6922 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6923 dpif_execute(ofproto->backer->dpif, key.data, key.size,
6924 odp_actions.data, odp_actions.size, packet);
6925 ofpbuf_uninit(&odp_actions);
6933 set_netflow(struct ofproto *ofproto_,
6934 const struct netflow_options *netflow_options)
6936 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6938 if (netflow_options) {
6939 if (!ofproto->netflow) {
6940 ofproto->netflow = netflow_create();
6942 return netflow_set_options(ofproto->netflow, netflow_options);
6944 netflow_destroy(ofproto->netflow);
6945 ofproto->netflow = NULL;
6951 get_netflow_ids(const struct ofproto *ofproto_,
6952 uint8_t *engine_type, uint8_t *engine_id)
6954 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6956 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
6960 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6962 if (!facet_is_controller_flow(facet) &&
6963 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6964 struct subfacet *subfacet;
6965 struct ofexpired expired;
6967 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6968 if (subfacet->path == SF_FAST_PATH) {
6969 struct dpif_flow_stats stats;
6971 subfacet_reinstall(subfacet, &stats);
6972 subfacet_update_stats(subfacet, &stats);
6976 expired.flow = facet->flow;
6977 expired.packet_count = facet->packet_count;
6978 expired.byte_count = facet->byte_count;
6979 expired.used = facet->used;
6980 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6985 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6987 struct facet *facet;
6989 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6990 send_active_timeout(ofproto, facet);
6994 static struct ofproto_dpif *
6995 ofproto_dpif_lookup(const char *name)
6997 struct ofproto_dpif *ofproto;
6999 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7000 hash_string(name, 0), &all_ofproto_dpifs) {
7001 if (!strcmp(ofproto->up.name, name)) {
7009 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7010 const char *argv[], void *aux OVS_UNUSED)
7012 struct ofproto_dpif *ofproto;
7015 ofproto = ofproto_dpif_lookup(argv[1]);
7017 unixctl_command_reply_error(conn, "no such bridge");
7020 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7022 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7023 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7027 unixctl_command_reply(conn, "table successfully flushed");
7031 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7032 const char *argv[], void *aux OVS_UNUSED)
7034 struct ds ds = DS_EMPTY_INITIALIZER;
7035 const struct ofproto_dpif *ofproto;
7036 const struct mac_entry *e;
7038 ofproto = ofproto_dpif_lookup(argv[1]);
7040 unixctl_command_reply_error(conn, "no such bridge");
7044 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7045 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7046 struct ofbundle *bundle = e->port.p;
7047 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7048 ofbundle_get_a_port(bundle)->odp_port,
7049 e->vlan, ETH_ADDR_ARGS(e->mac),
7050 mac_entry_age(ofproto->ml, e));
7052 unixctl_command_reply(conn, ds_cstr(&ds));
7057 struct action_xlate_ctx ctx;
7063 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7064 const struct rule_dpif *rule)
7066 ds_put_char_multiple(result, '\t', level);
7068 ds_put_cstr(result, "No match\n");
7072 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7073 table_id, ntohll(rule->up.flow_cookie));
7074 cls_rule_format(&rule->up.cr, result);
7075 ds_put_char(result, '\n');
7077 ds_put_char_multiple(result, '\t', level);
7078 ds_put_cstr(result, "OpenFlow ");
7079 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7080 ds_put_char(result, '\n');
7084 trace_format_flow(struct ds *result, int level, const char *title,
7085 struct trace_ctx *trace)
7087 ds_put_char_multiple(result, '\t', level);
7088 ds_put_format(result, "%s: ", title);
7089 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7090 ds_put_cstr(result, "unchanged");
7092 flow_format(result, &trace->ctx.flow);
7093 trace->flow = trace->ctx.flow;
7095 ds_put_char(result, '\n');
7099 trace_format_regs(struct ds *result, int level, const char *title,
7100 struct trace_ctx *trace)
7104 ds_put_char_multiple(result, '\t', level);
7105 ds_put_format(result, "%s:", title);
7106 for (i = 0; i < FLOW_N_REGS; i++) {
7107 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7109 ds_put_char(result, '\n');
7113 trace_format_odp(struct ds *result, int level, const char *title,
7114 struct trace_ctx *trace)
7116 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7118 ds_put_char_multiple(result, '\t', level);
7119 ds_put_format(result, "%s: ", title);
7120 format_odp_actions(result, odp_actions->data, odp_actions->size);
7121 ds_put_char(result, '\n');
7125 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7127 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7128 struct ds *result = trace->result;
7130 ds_put_char(result, '\n');
7131 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7132 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7133 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7134 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7138 trace_report(struct action_xlate_ctx *ctx, const char *s)
7140 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7141 struct ds *result = trace->result;
7143 ds_put_char_multiple(result, '\t', ctx->recurse);
7144 ds_put_cstr(result, s);
7145 ds_put_char(result, '\n');
7149 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7150 void *aux OVS_UNUSED)
7152 const char *dpname = argv[1];
7153 struct ofproto_dpif *ofproto;
7154 struct ofpbuf odp_key;
7155 struct ofpbuf *packet;
7156 ovs_be16 initial_tci;
7162 ofpbuf_init(&odp_key, 0);
7165 ofproto = ofproto_dpif_lookup(dpname);
7167 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7171 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7172 /* ofproto/trace dpname flow [-generate] */
7173 const char *flow_s = argv[2];
7174 const char *generate_s = argv[3];
7176 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7177 * flow. We guess which type it is based on whether 'flow_s' contains
7178 * an '(', since a datapath flow always contains '(') but an
7179 * OpenFlow-like flow should not (in fact it's allowed but I believe
7180 * that's not documented anywhere).
7182 * An alternative would be to try to parse 'flow_s' both ways, but then
7183 * it would be tricky giving a sensible error message. After all, do
7184 * you just say "syntax error" or do you present both error messages?
7185 * Both choices seem lousy. */
7186 if (strchr(flow_s, '(')) {
7187 enum odp_key_fitness fitness;
7190 /* Convert string to datapath key. */
7191 ofpbuf_init(&odp_key, 0);
7192 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7194 unixctl_command_reply_error(conn, "Bad flow syntax");
7198 fitness = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
7199 flow.in_port = odp_port_to_ofp_port(ofproto, flow.in_port);
7201 /* Convert odp_key to flow. */
7202 error = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
7203 &initial_tci, NULL);
7204 if (error == ODP_FIT_ERROR) {
7205 unixctl_command_reply_error(conn, "Invalid flow");
7211 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7213 unixctl_command_reply_error(conn, error_s);
7218 initial_tci = flow.vlan_tci;
7219 vsp_adjust_flow(ofproto, &flow);
7222 /* Generate a packet, if requested. */
7224 packet = ofpbuf_new(0);
7225 flow_compose(packet, &flow);
7227 } else if (argc == 7) {
7228 /* ofproto/trace dpname priority tun_id in_port mark packet */
7229 const char *priority_s = argv[2];
7230 const char *tun_id_s = argv[3];
7231 const char *in_port_s = argv[4];
7232 const char *mark_s = argv[5];
7233 const char *packet_s = argv[6];
7234 uint32_t in_port = atoi(in_port_s);
7235 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7236 uint32_t priority = atoi(priority_s);
7237 uint32_t mark = atoi(mark_s);
7240 msg = eth_from_hex(packet_s, &packet);
7242 unixctl_command_reply_error(conn, msg);
7246 ds_put_cstr(&result, "Packet: ");
7247 s = ofp_packet_to_string(packet->data, packet->size);
7248 ds_put_cstr(&result, s);
7251 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7252 flow.tunnel.tun_id = tun_id;
7253 initial_tci = flow.vlan_tci;
7255 unixctl_command_reply_error(conn, "Bad command syntax");
7259 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7260 unixctl_command_reply(conn, ds_cstr(&result));
7263 ds_destroy(&result);
7264 ofpbuf_delete(packet);
7265 ofpbuf_uninit(&odp_key);
7269 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7270 const struct ofpbuf *packet, ovs_be16 initial_tci,
7273 struct rule_dpif *rule;
7275 ds_put_cstr(ds, "Flow: ");
7276 flow_format(ds, flow);
7277 ds_put_char(ds, '\n');
7279 rule = rule_dpif_lookup(ofproto, flow);
7281 trace_format_rule(ds, 0, 0, rule);
7282 if (rule == ofproto->miss_rule) {
7283 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7284 } else if (rule == ofproto->no_packet_in_rule) {
7285 ds_put_cstr(ds, "\nNo match, packets dropped because "
7286 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7290 uint64_t odp_actions_stub[1024 / 8];
7291 struct ofpbuf odp_actions;
7293 struct trace_ctx trace;
7296 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7299 ofpbuf_use_stub(&odp_actions,
7300 odp_actions_stub, sizeof odp_actions_stub);
7301 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7302 rule, tcp_flags, packet);
7303 trace.ctx.resubmit_hook = trace_resubmit;
7304 trace.ctx.report_hook = trace_report;
7305 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7308 ds_put_char(ds, '\n');
7309 trace_format_flow(ds, 0, "Final flow", &trace);
7310 ds_put_cstr(ds, "Datapath actions: ");
7311 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7312 ofpbuf_uninit(&odp_actions);
7314 if (trace.ctx.slow) {
7315 enum slow_path_reason slow;
7317 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7318 "slow path because it:");
7319 for (slow = trace.ctx.slow; slow; ) {
7320 enum slow_path_reason bit = rightmost_1bit(slow);
7324 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7327 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7330 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7333 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7336 ds_put_cstr(ds, "\n\t (The datapath actions are "
7337 "incomplete--for complete actions, "
7338 "please supply a packet.)");
7341 case SLOW_CONTROLLER:
7342 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7343 "to the OpenFlow controller.");
7346 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7347 "than the datapath supports.");
7354 if (slow & ~SLOW_MATCH) {
7355 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7356 "the special slow-path processing.");
7363 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7364 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7367 unixctl_command_reply(conn, NULL);
7371 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7372 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7375 unixctl_command_reply(conn, NULL);
7378 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7379 * 'reply' describing the results. */
7381 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7383 struct facet *facet;
7387 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7388 if (!facet_check_consistency(facet)) {
7393 ofproto->need_revalidate = REV_INCONSISTENCY;
7397 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7398 ofproto->up.name, errors);
7400 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7405 ofproto_dpif_self_check(struct unixctl_conn *conn,
7406 int argc, const char *argv[], void *aux OVS_UNUSED)
7408 struct ds reply = DS_EMPTY_INITIALIZER;
7409 struct ofproto_dpif *ofproto;
7412 ofproto = ofproto_dpif_lookup(argv[1]);
7414 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7415 "ofproto/list for help)");
7418 ofproto_dpif_self_check__(ofproto, &reply);
7420 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7421 ofproto_dpif_self_check__(ofproto, &reply);
7425 unixctl_command_reply(conn, ds_cstr(&reply));
7429 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7430 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7431 * to destroy 'ofproto_shash' and free the returned value. */
7432 static const struct shash_node **
7433 get_ofprotos(struct shash *ofproto_shash)
7435 const struct ofproto_dpif *ofproto;
7437 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7438 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7439 shash_add_nocopy(ofproto_shash, name, ofproto);
7442 return shash_sort(ofproto_shash);
7446 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7447 const char *argv[] OVS_UNUSED,
7448 void *aux OVS_UNUSED)
7450 struct ds ds = DS_EMPTY_INITIALIZER;
7451 struct shash ofproto_shash;
7452 const struct shash_node **sorted_ofprotos;
7455 shash_init(&ofproto_shash);
7456 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7457 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7458 const struct shash_node *node = sorted_ofprotos[i];
7459 ds_put_format(&ds, "%s\n", node->name);
7462 shash_destroy(&ofproto_shash);
7463 free(sorted_ofprotos);
7465 unixctl_command_reply(conn, ds_cstr(&ds));
7470 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7472 struct dpif_dp_stats s;
7473 const struct shash_node **ports;
7476 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7478 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7479 dpif_name(ofproto->backer->dpif));
7480 /* xxx It would be better to show bridge-specific stats instead
7481 * xxx of dp ones. */
7483 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7484 s.n_hit, s.n_missed, s.n_lost);
7485 ds_put_format(ds, "\tflows: %zu\n",
7486 hmap_count(&ofproto->subfacets));
7488 ports = shash_sort(&ofproto->up.port_by_name);
7489 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7490 const struct shash_node *node = ports[i];
7491 struct ofport *ofport = node->data;
7492 const char *name = netdev_get_name(ofport->netdev);
7493 const char *type = netdev_get_type(ofport->netdev);
7495 ds_put_format(ds, "\t%s %u/%u:", name, ofport->ofp_port,
7496 ofp_port_to_odp_port(ofproto, ofport->ofp_port));
7497 if (strcmp(type, "system")) {
7498 struct netdev *netdev;
7501 ds_put_format(ds, " (%s", type);
7503 error = netdev_open(name, type, &netdev);
7508 error = netdev_get_config(netdev, &config);
7510 const struct smap_node **nodes;
7513 nodes = smap_sort(&config);
7514 for (i = 0; i < smap_count(&config); i++) {
7515 const struct smap_node *node = nodes[i];
7516 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7517 node->key, node->value);
7521 smap_destroy(&config);
7523 netdev_close(netdev);
7525 ds_put_char(ds, ')');
7527 ds_put_char(ds, '\n');
7533 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7534 const char *argv[], void *aux OVS_UNUSED)
7536 struct ds ds = DS_EMPTY_INITIALIZER;
7537 const struct ofproto_dpif *ofproto;
7541 for (i = 1; i < argc; i++) {
7542 ofproto = ofproto_dpif_lookup(argv[i]);
7544 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7545 "for help)", argv[i]);
7546 unixctl_command_reply_error(conn, ds_cstr(&ds));
7549 show_dp_format(ofproto, &ds);
7552 struct shash ofproto_shash;
7553 const struct shash_node **sorted_ofprotos;
7556 shash_init(&ofproto_shash);
7557 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7558 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7559 const struct shash_node *node = sorted_ofprotos[i];
7560 show_dp_format(node->data, &ds);
7563 shash_destroy(&ofproto_shash);
7564 free(sorted_ofprotos);
7567 unixctl_command_reply(conn, ds_cstr(&ds));
7572 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
7573 int argc OVS_UNUSED, const char *argv[],
7574 void *aux OVS_UNUSED)
7576 struct ds ds = DS_EMPTY_INITIALIZER;
7577 const struct ofproto_dpif *ofproto;
7578 struct subfacet *subfacet;
7580 ofproto = ofproto_dpif_lookup(argv[1]);
7582 unixctl_command_reply_error(conn, "no such bridge");
7586 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
7587 struct odputil_keybuf keybuf;
7590 subfacet_get_key(subfacet, &keybuf, &key);
7591 odp_flow_key_format(key.data, key.size, &ds);
7593 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
7594 subfacet->dp_packet_count, subfacet->dp_byte_count);
7595 if (subfacet->used) {
7596 ds_put_format(&ds, "%.3fs",
7597 (time_msec() - subfacet->used) / 1000.0);
7599 ds_put_format(&ds, "never");
7601 if (subfacet->facet->tcp_flags) {
7602 ds_put_cstr(&ds, ", flags:");
7603 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
7606 ds_put_cstr(&ds, ", actions:");
7607 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
7608 ds_put_char(&ds, '\n');
7611 unixctl_command_reply(conn, ds_cstr(&ds));
7616 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
7617 int argc OVS_UNUSED, const char *argv[],
7618 void *aux OVS_UNUSED)
7620 struct ds ds = DS_EMPTY_INITIALIZER;
7621 struct ofproto_dpif *ofproto;
7623 ofproto = ofproto_dpif_lookup(argv[1]);
7625 unixctl_command_reply_error(conn, "no such bridge");
7629 flush(&ofproto->up);
7631 unixctl_command_reply(conn, ds_cstr(&ds));
7636 ofproto_dpif_unixctl_init(void)
7638 static bool registered;
7644 unixctl_command_register(
7646 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
7647 2, 6, ofproto_unixctl_trace, NULL);
7648 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7649 ofproto_unixctl_fdb_flush, NULL);
7650 unixctl_command_register("fdb/show", "bridge", 1, 1,
7651 ofproto_unixctl_fdb_show, NULL);
7652 unixctl_command_register("ofproto/clog", "", 0, 0,
7653 ofproto_dpif_clog, NULL);
7654 unixctl_command_register("ofproto/unclog", "", 0, 0,
7655 ofproto_dpif_unclog, NULL);
7656 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7657 ofproto_dpif_self_check, NULL);
7658 unixctl_command_register("dpif/dump-dps", "", 0, 0,
7659 ofproto_unixctl_dpif_dump_dps, NULL);
7660 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
7661 ofproto_unixctl_dpif_show, NULL);
7662 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
7663 ofproto_unixctl_dpif_dump_flows, NULL);
7664 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
7665 ofproto_unixctl_dpif_del_flows, NULL);
7668 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7670 * This is deprecated. It is only for compatibility with broken device drivers
7671 * in old versions of Linux that do not properly support VLANs when VLAN
7672 * devices are not used. When broken device drivers are no longer in
7673 * widespread use, we will delete these interfaces. */
7676 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7678 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7679 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7681 if (realdev_ofp_port == ofport->realdev_ofp_port
7682 && vid == ofport->vlandev_vid) {
7686 ofproto->need_revalidate = REV_RECONFIGURE;
7688 if (ofport->realdev_ofp_port) {
7691 if (realdev_ofp_port && ofport->bundle) {
7692 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7693 * themselves be part of a bundle. */
7694 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7697 ofport->realdev_ofp_port = realdev_ofp_port;
7698 ofport->vlandev_vid = vid;
7700 if (realdev_ofp_port) {
7701 vsp_add(ofport, realdev_ofp_port, vid);
7708 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7710 return hash_2words(realdev_ofp_port, vid);
7713 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7714 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7715 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7716 * it would return the port number of eth0.9.
7718 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7719 * function just returns its 'realdev_odp_port' argument. */
7721 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7722 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7724 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7725 uint16_t realdev_ofp_port;
7726 int vid = vlan_tci_to_vid(vlan_tci);
7727 const struct vlan_splinter *vsp;
7729 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
7730 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7731 hash_realdev_vid(realdev_ofp_port, vid),
7732 &ofproto->realdev_vid_map) {
7733 if (vsp->realdev_ofp_port == realdev_ofp_port
7734 && vsp->vid == vid) {
7735 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
7739 return realdev_odp_port;
7742 static struct vlan_splinter *
7743 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7745 struct vlan_splinter *vsp;
7747 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7748 &ofproto->vlandev_map) {
7749 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7757 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7758 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7759 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7760 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7761 * eth0 and store 9 in '*vid'.
7763 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7764 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7767 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7768 uint16_t vlandev_ofp_port, int *vid)
7770 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7771 const struct vlan_splinter *vsp;
7773 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7778 return vsp->realdev_ofp_port;
7784 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7785 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7786 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7787 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7788 * always the case unless VLAN splinters are enabled), returns false without
7789 * making any changes. */
7791 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7796 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7801 /* Cause the flow to be processed as if it came in on the real device with
7802 * the VLAN device's VLAN ID. */
7803 flow->in_port = realdev;
7804 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7809 vsp_remove(struct ofport_dpif *port)
7811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7812 struct vlan_splinter *vsp;
7814 vsp = vlandev_find(ofproto, port->up.ofp_port);
7816 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7817 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7820 port->realdev_ofp_port = 0;
7822 VLOG_ERR("missing vlan device record");
7827 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7829 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7831 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7832 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7833 == realdev_ofp_port)) {
7834 struct vlan_splinter *vsp;
7836 vsp = xmalloc(sizeof *vsp);
7837 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7838 hash_int(port->up.ofp_port, 0));
7839 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7840 hash_realdev_vid(realdev_ofp_port, vid));
7841 vsp->realdev_ofp_port = realdev_ofp_port;
7842 vsp->vlandev_ofp_port = port->up.ofp_port;
7845 port->realdev_ofp_port = realdev_ofp_port;
7847 VLOG_ERR("duplicate vlan device record");
7852 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
7854 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
7855 return ofport ? ofport->odp_port : OVSP_NONE;
7858 static struct ofport_dpif *
7859 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
7861 struct ofport_dpif *port;
7863 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
7864 hash_int(odp_port, 0),
7865 &backer->odp_to_ofport_map) {
7866 if (port->odp_port == odp_port) {
7875 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
7877 struct ofport_dpif *port;
7879 port = odp_port_to_ofport(ofproto->backer, odp_port);
7880 if (port && ofproto == ofproto_dpif_cast(port->up.ofproto)) {
7881 return port->up.ofp_port;
7887 const struct ofproto_class ofproto_dpif_class = {
7922 port_is_lacp_current,
7923 NULL, /* rule_choose_table */
7930 rule_modify_actions,
7939 get_cfm_remote_mpids,
7944 get_stp_port_status,
7951 is_mirror_output_bundle,
7952 forward_bpdu_changed,
7953 set_mac_table_config,